[This package may not work! Using Gtk.jl might just be best...]

A package to simplify the creation of GUIs within Julia

The JGUI package provides a few different means to ease the creation of GUIs within Julia. These include a simplified implementation of Mathematica's Manipulate function, and a simplified interface for using the tcl/tk, Gtk, or Qt toolkits within Julia.

The JGUI package installs with Pkg.clone("https://github.com/jverzani/JGUI.jl"). For it to work one needs to have installed the Tk package, the Gtk package, or the PySide package. The latter requires an installation of the Qt libraries (http://qt-project.org/downloads), Python (http://www.python.org/download/), the PySide (http://qt-project.org/wiki/Get-PySide) interface between Python and Qt, and the PyCall package to connect Python with Julia (installed with the PySide package). The Anaconda (http://docs.continuum.io/anaconda/) packaging of Python should be a one-stop installation, though the Qt part can be buggy.

The easiest way to create a GUI with this package is to use the manipulate function, which can be used to evaluate an expression parameterized by values coming from easily specified controls within a GUI.

The first line below allows one to specify the toolkit, currently Tk, Gtk (default) or Qt. Mixing and matching within a Julia session will likely lead to crashes. The toolkit is specified before JGUI is loaded, so that toolkit-specific code can be loaded.

ENV["toolkit"] = "Gtk"		# default, can be skipped
using JGUI

Now, consider the following expression which when evaluated produces a Winston plot object:

expr = quote
     using Winston
     plot(x -> sin(u*x), 0, 2pi)
end

We can use the manipulate function to fill in values for the unbound variable u, when the expression is evaluated. The plot command above returns a Winston plot object, which is then plotted. To create a control to specify values for u, we simply need to specify a range, as follows:

a = manipulate(expr, (:u, 1:10))

This call will pop up a simple GUI with a slider that allows one to adjust the value of u from 1 to 10, updating the graphic as this is done.

Here is a how one can add a title to the plot. First we modify the plot call to include a title:

expr = quote
     using Winston
     xs = linspace(0, 2pi)
     ys = [sin(u*x) for x in xs]
     plot(x, y, title=title)
end

Now title is also unbound. To specify a control to set a title, we use a string:

a = manipulate(expr, (:u, 1:10), (:title, "A sine plot"))

Now when the plot is updated, the title is also taken from a text box.

Manipulate has other simple-to-specify controls:

• (:symbol, Bool) - checkbox. Use {:label=>"some label"} to label it.
• (:symbol, Range) - slider
• (:symbol, Range, Range) - 2d slider
• (:symbol, Vector) - radio or combobox (depends on size)
• (:symbol, String) - text edit
• (:symbol, Real) - text edit with conversion to float via parsefloat
• (:symbol, Int) - text edit with conversion to integer via parseint

The expression can evaluate to a Winston plot object or any other object. Plot objects are rendered in a graphics device, text objects in a text box.

When using Qt (ENV["toolkit"] = "Qt") one can plot PyPlot calls, not Winston calls. The modules argument should include PyPlot.

The Gtk, Tk, and PySide packages for Julia provide relatively easy means to produce GUIs with the Gtk, Tcl/Tk and Qt toolkits, respectively. In addition to the manipulate function, this package provides a small API for creating GUIs that makes it even easier than the other packages, though sacrificing a huge amount of their power and flexibility. (The JGUI interface is primarily concerned with simpler things like controls, and not more involved interfaces like those with a canvas widget.)

Here is a simple example where a window has a button which when clicked will destroy the window.

w = window(size=[200, 200])
w[:title] = "hello world"

b = button(w, "Close")
push!(w, b)

connect(b, "clicked", w, destroy)
raise(w)

The first line creates a window object with an optional size specified.

The second line shows how a property of the window object may be set, using indexing notation with an appropriate symbol, in this case :title. There are relatively few properties for any given object. For a control, the most important is :value. (The method properties will list all of a widget's properties.)

The third line creates a button object. All constructors except window use a parent container for the first argument. (This is similar to Tk and so the widget hierarchy is determined, but not the actual layout). The button constructor has label value for the second positional argument.

The fourth line is specific to JGUI and not the underlying toolkit. (Though, the same idea is present in Gtk.) Rather than use separate layout managers, as is done with Qt or Tk, each container object is conceptualized as a queue of some sort. For the window object, the push! method adds the button to the window queue, laying it out as it does so.

The fifth line is how one connects a callback to an event. In this case the receiver of a click event, b, will emit a signal clicked. The w object is passed to its method destroy to destroy the window. This pattern follows Qt's signal-and-slots style. One can also just pass in a function to call instead of the last two arguments, something like connect(b, "clicked", () -> destroy(w)).

Finally, the window is raised.

Though simple, the above example demonstrates most all the procedures when creating a GUI: creating GUI objects, accessing their properties, laying out the objects, and creating interactivity by assigning callbacks to user-initiated events.

Let's look at another example, this one mimics, the first manipulate example.

## needs Gtk or Tk
ENV["toolkit"] = "Gtk"
using JGUI, Winston

w = window()
f = hbox(w); push!(f)

sl = slider(f, 1:10)
sl[:size] = [100, 20]
cnv = cairographic(f)

append!(f, [sl, cnv])

connect(sl, "valueChanged") do u
  p = plot(x -> sin(u * x), 0, 2pi)
  display(cnv, p)
end

notify(sl, "valueChanged", 1) # draw initial graphic

In the above we have several constructors: window, hbox, slider, and cairographic. Each produces a widget. The window creates a toplevel window, and slider a slider. The hbox constructor creates a horizontal box container, which is used above to hold two children, the slider and a cairo graphic device produced by cairographic. (The pyplotgraphic widget produces a device for graphics drawn via PyPlot and can be use with the Qt toolkit.)

As mentioned, constructors in JGUI, except for window, have a parent container passed as the first argument. Additional arguments are used to customize the constructor. For hbox and cairographic, there is no needed customization, though the latter may have a width and height argument specified. For a slider, one needs to specify the range that is stepped over. Unlike most slider implementations, such as the one in Tk, a slider is used to select amongst the specified range or sorted vector. This reduces the need to specify a step size and is more in line with how julia produces sequences of values.

For a slider, the valueChanged signal passes the new value to the callback. This value is then used within the callback that produces the graphic. One could also access this value within the callback with sl[:value].

The cairographic widget is not integrated into the multimedia display system (out of ignorance). Following Winston, the display method above takes as a first argument the cairographic widget and the second object a Winston object to plot. If doing this at the command line, use a semicolon after the plot call to avoid displaying automatically.

The last line is one hacky way to get the initial graphic drawn. The notify method of the underlying model notifies any observers of a "valueChanged" event for a specified value.

Constructors produce basically two types of objects: controls and containers. The containers available in this package are few:

• hbox and vbox produce horizontal and vertical box containers.

• grid makes a container for arranging its children in a grid.

• notebook provides a tabbed notebook for organizing its children.

• formlayout provides a simple way to lay out label/controls in a grid

• labelframe provides a simple container for holding a single child (like window), in this case with a label and decorative frame.

Containers are coupled with a layout manager which are utilized in a "julian" manner:

• The hbox and vbox containers have methods push!, insert!, and append! for adding children to the layout; pop! and splice! for removing children. In the above example, we use append! to add two children at a time.

• The formlayout and notebook containers also implement the above for adding a child at a time, with an additional label.

• The single-child containers, labelframe and window, use push! to add their child.

• Children of a grid container are managed via matrix notation. There are two styles. One can add a matrix of widgets:

w = window(title="Matrix of widgets")
g = grid(w); push!(g)		# push!(g) is same as push!(w, g)
b1 = button(g, "one")
b2 = button(g, "two")
b3 = button(g, "three")

g[:,:] = [b1 b2; nothing b3]
raise(w)

Or one can add a single child using [row,column] notation. These may be specified through a range to span multiple rows or columns.

The expanding and alignment properties of how a child is placed into a parent are specified for the child, not the container. These are done through the properties :sizepolicy and :alignment. Padding is done through the :spacing properties of the containers.

Widgets have properties that can be queried and set through index notation where a symbol is used for indexing. For example, to set the size policy of a widget, we have:

w = window(size=[300, 300])
f = hbox(w); push!(f)
b = button(f, "expanding")

#b[:sizepolicy] = (:expand, :fixed)  # expand in x direction
b[:sizepolicy] = (:fixed, :expand)   # expand in y direction
#b[:sizepolicy] = (:expand, :expand) # expand in both

push!(b)
raise(w)

Some properties are dynamic, this one is not. It should be set before packing into a layout.

The main value of a widget is assigned the value property. For a button this is its label:

w = window(title="change label")
b = button(w, "old label"); push!(b)
b[:value]			# "old label"
b[:value] = "new label"		# updates button
b[:value]			# "new label"

When a property, say :prop, is looked up a search for either a getProp or setProp method is made. Though not exported, save for getValue and setValue, these functions can be conveniently employed when using the property in a callback.

The basic connect method is used to connect a callback to an event. The syntax follows Qt's signals and slots usage. It can take two forms: connect(receiver, signal, obj, slot) or connect(receiver, signal, slot), where slot is a function. In the first instance, the call is slot(obj, vals...) and the second, just slot(vals...) where vals... depends on the signal: the basic valueChanged signal passes in the value; whereas, a button's clicked signal has no value passed. One can also use the do syntax to specify the slot as a function.

Widgets have different signals defined. Mostly the names follow a small subset of those for the corresponding Qt widget (hence the names in camelCase format).

The connect method returns an id. This can be used with disconnect to remove an observer of an object. At present there is no way to temporarily suspend a callback.

As an example, This is how one connects a slider value to a label:

w = window(title="label and slider")
f = hbox(w); push!(f)
sl = slider(f, 1:20)
sl[:size] = [100, 20]
l = label(f, sl[:value])
append!(f, [sl, l])

connect(sl, "valueChanged", l, setValue)
raise(w)

Some alternatives would be connect(rb, "valueChanged", l, (l, value) -> l[:value] = value) or connect(rb, "valueChanged", value -> l[:value] = value).

As an aside, this can also be done just by sharing the underlying model, as with:

w = window(title="label and slider")
f = hbox(w); push!(f)
sl = slider(f, 1:20)
sl[:size] = [100, 20]
l = label(f, sl.model)
append!(f, [sl, l])

The React.jl package is a signaling package. It will be integrated more fully, but for the output of one widget can be used as the input of another. The above example can be done via:

w =  window(title="label and slider")
f = hbox(w); push!(f)
sl = slider(f, 1:20)
sl[:size] = [100, 20]
l = label(f, "")
l[:value] = sl			# connect value of slider to value of label.
append!(f, [sl, l])
raise(w)

The basic widgets are:

• label a standard text label

• separator used to place a horizontal or vertical line in a layout

• button a push button

• lineedit a single line text edit

• textedit multi-line text edit

• checkbox a simple true/false toggle

• radiogroup exclusive set of buttons

• buttongroup exclusive (or not) set of buttons

• combobox a popup selection widget

• slider select from numeric range

• slider2d select two variables from numeric range. (Not Gtk.)

• listview Show a vector of values allowing selection of one or more. (Not Gtk.)

• storeview used to display store of records.

• treeview used to display tree structured records. (Not Gtk.)

• cairographic used with Winston graphics (Gtk and Tk only)

• pyplotgraphic used with PyPlot graphics (Qt only)

• imageview used to display png or gif image files.

The cairographic widget is a light wrapper around a Cairo canvas provided by the Tk and Gtk packages.

To use the canvas, access the :widget property of the cairographic object:

## update two graphics windows...
ENV["toolkit"] = "Gtk"
using JGUI, Winston
w = window()
f = grid(w); push!(f)
cnv1 = cairographic(f, width=480, height=400)
cnv2 = cairographic(f, width=480, height=400)
b = button(f, "update"); b[:alignment] = (:right, :center)
f[:,:] = [cnv1 cnv2; nothing b]
connect(b, "clicked") do
   p = FramedPlot(); add(p, Curve(rand(10), rand(10))); display(cnv1, p)
   p = FramedPlot(); add(p, Curve(rand(10), rand(10))); display(cnv2, p)
end
notify(b, "clicked")	# roundabout way to draw initial graphic, ...

A common widget is a tabular display of data, where each row represents a case, and column some measurement associate to that case. Basically a spreadsheet with some consistency in the usage. The storeview widget can display such data.

A Store is a vector of records, or cases. Each record is an array of type Any or a tuple. The records are displayed in a grid , one row per record. The storeview widget is used to display values in an underlying Store object.

Here is an example. First a store is defined. We specify the type of variables for each column, as this information is needed in some toolkits (such as Gtk, where this design is used for @GtkListStore).

store = Store(Int, Float64, String)
push!(store, (1, 1.0, "one"))
push!(store, (2, 2.0, "two"))
push!(store, (3, 3.0, "three"))

Here is how we lay it out:

w = window(size=[300, 300])
sv = storeview(w, store)
push!(sv)			
sv[:names] = ["Int", "Float64", "String"]
sv[:widths] = [100, 100, 100]	# column widths
sv[:selectmode] = :multiple	# :single or :multiple
id = connect(sv, "clicked", (row, col) -> println((row, col))) # sample handler

raise(w)

One can add and remove items through insert!, splice!; one can modify existing items through indexing:

t4 = (4, 4.0, "four")
push!(store, t4)
splice!(store, 1)
item = store[1]
item[3] = uppercase(item[3])
store[1] = item

The valueChanged signal passes the index (or indices) that are selected. These are also given by the :index property. The :value property returns the items. One uses :index to set the selection, not :value.

In addition to rowClicked, there are rowDoubleClicked, headerClicked, and selectionChanged signals.

A treeview uses a treestore to hold the data. Here is a simple example:

ENV["toolkit"] = "Tk"		# not Gtk!!!

tstore = treestore(Int, Float64, String)

t1 = (1, 1.0, "one")
t11 = (11, 11.0, "one-one")
t2  = (2, 2.0, "two")


w = window(size=[300, 300])
tv = treeview(w, tstore)
tv[:names] = ["Int", "Float", "String"]
tv[:widths] = [50,50,-1]

tv[:keyname] = "key"
tv[:keywidth] = 100

push!(w, tv)  

raise(w)

To manage child items, we have insert! (with signature (store, parent, sibling position, label, [data])):

node = insert!(tstore, nothing, 1, "label1", t1)
insert!(tstore, node, 1, "label11", t11)
node = insert!(tstore, nothing, 2, "label2", t2)

Nodes are related to a path, which specifies the ancestry. The path [3,2,1] would be the first child of the second child of the third child of the root. We use the path to find a node to open via:

node = path_to_node(tstore, [1])
expand_node(tv, node)

We can remove nodes via a two-argument form of pop!

node = path_to_node(tstore, [1,1])
pop!(tstore, node)

There are some standard modal dialogs

• filedialog

• messagebox

• confirmbox

In addition, the dialog constructor can be used to generate dialogs, somewhat similar to Qt's base Dialog class:

ENV["toolkit"] = "Tk"		# not Gtk!!!
using JGUI
w = window()			# Some parent to position the dialog near

dlg = dialog(w, buttons=[:cancel, :ok]) # default is just `:ok`
f = vbox(dlg); push!(f)

l = label(f, "More complicated controls go here"); push!(l)

connect(dlg, "finished", state -> println(state))
dlg.exec()