When the C++ binding of gtk+, gtkmm, deletes the C++ wrapper of a widget, it
calls g_object_run_dispose(object), where 'object' is the wrapped gtk+ widget.
The call to g_object_run_dispose() makes the widget disconnect from its parent
by calling gtk_container_remove().

This usually works fine, and has done so for many years. But now it causes
segfault in assistant_remove_page_cb(), if one of GtkAssistant's pages is
disposed of. That's because of the commit
https://git.gnome.org/browse/gtk+/commit/?id=6cc65260f4ba54ea0ca9f93068050435bd6b9f1a

Previously, the parent of a page was the notebook in the assistant. When a page
widget removed itself from its parent, it was removed from the notebook. Then
assistant_remove_page_cb() was called.

Now the parent of a page is a box, created by gtk_assistant_insert_page().
When a page widget removes itself from its parent, it is removed from the box.
Then assistant_remove_page_cb() is *not* called.
Later, when assistant_remove_page_cb() is called because the box is removed
from the notebook, the box has no children. The program crashes because
children is NULL in the following code, copied from assistant_remove_page_cb().

  if (GTK_IS_BOX (page))
    {
      GList *children;

      children = gtk_container_get_children (GTK_CONTAINER (page));
      page = GTK_WIDGET (children->data);
      g_list_free (children);
    }

It would be easy to test if children is NULL after the call to
gtk_container_get_children(), but I suppose that's not enough.
assistant_remove_page_cb() should be called when a page is removed from its
parent.