How to modify the navigation toolbar easily in a matplotlib figure window?

Question

Is it possible to do something like the following to modify the navigation toolbar in matplotlib?

1. Generate a figure window, with: fig = figure()
2. Get a reference of the navigation tool-bar, with: tbar = fig.get_navigation_toolbar(), or better yet, just by: tbar = fig.navtbar
3. Modify the tool-bar through the reference tbar, such as delete/add/edit a button with something like this:
      tbar.add_button(<a Button object>);
   tbar.remove_button(a reference to a button);
   tbar.edit_button(a reference to a button);
   
4. Update the figure with: fig.canvas.draw()

Thank you very much.

Answer 1

The way I found to remove unwanted toolbar items is making a subclass, which is instantiated and used in a GTK application. As I manually create Figure, FigureCanvas and NavigationToolbar objects anyway, this was the easiest way.

class NavigationToolbar(NavigationToolbar2GTKAgg):
    # only display the buttons we need
    toolitems = [t for t in NavigationToolbar2GTKAgg.toolitems if
                 t[0] in ('Home', 'Pan', 'Zoom', 'Save')]

If you want to create custom buttons, you should take a look on the definition of NavigationToolbar2 in backend_bases. You can easily add your own entries to the toolitems list and define appropriate callback functions in your toolbar subclass.

Answer 2

With MPL 1.2.1 it is possible to get an handler of the navigation toolbar of a standard MPL figure through figure.canvas.toolbar. I'm not sure about previous versions.

At least with the QT backend it is possible to add arbitrary widgets to the navigation toolbar using the QT method .addWidget(). I suppose other backends will work using similar methods, but I haven't tested them.

Here it is a working example (using the QT backend) that adds a QLineEdit() to the navigation toolbar to change the title of a MPL figure (run from IPython (pylab) with run -i ..., then launch test()):

from PySide import QtGui, QtCore

def test():
    plot([1,2,3], lw=2)
    q = qt4_interface(gcf())
    return q   # WARNING: it's paramount to return the object otherwise, with 
               # no references, python deletes it and the GUI doesn't respond!

class qt4_interface:
    def __init__(self,fig):
        self.fig = fig

        toolbar = fig.canvas.toolbar
        self.line_edit = QtGui.QLineEdit()
        toolbar.addWidget(self.line_edit)
        self.line_edit.editingFinished.connect(self.do_something) 

    def do_something(self, *args):
        self.fig.axes[0].set_title(self.line_edit.text())
        self.fig.canvas.draw()
        #f = open('l','a'); f.write('yes\n'); f.flush(); f.close()

Answer 3

The previous answers work but are very backend-specific. A slighly more elegant solution is to subclass NavigationToolbar2, as done in this other answer: Matplotlib/Tkinter - customizing toolbar tooltips There the aim was to change the tooltips, but adding or removing a button is equally trivial.