Hot Reload in Uno and GTK


While playing around with Microsoft’s Uno Platform, I discovered its super-neat XAML Hot Reload feature. It basically does exactly what you think it would; while the app is running, changes made to any XAML file will be reflected automatically, without needing to re-build anything. This is basically the desktop development equivalent of LiveReload, and is a great way to tighten the feedback loop and enable faster development.

Unfortunately, my laptop only runs Linux, and their documentation requires you to use the Visual Studio Add-in in order to use hot reload. After some searching, however, I discovered Matheus Castello’s post about getting hot reload to work on embedded Linux. This was very promising, but he did not go into any detail about how he got it working, only providing some videos of his Visual Studio Code extension.

However, after some work, I was able to get it working in an editor-agnostic way. Here’s Uno Hot Reload in action using nothing but tmux and Neovim:

The bottom-left pane is running the Uno Remote Control Host, which lives within Uno’s source tree, using something like this:

# As of writing, the Remote Control Host requires .NET Core 3.
# When using asdf with the dotnet-core plugin, you can set the correct
# version with:
$ asdf install dotnet-core 3.1.401 && asdf shell dotnet-core 3.1.401
$ source ~/.asdf/plugins/dotnet-core/set-dotnet-home.bash

# Now build and run the Remote Control Host.
# "${uno}" refers to the path of the Uno source tree.
$ cd "${uno}"/src/Uno.UI.RemoteControl.Host
$ "${DOTNET_ROOT}"/dotnet build
$ cd bin/Debug/netcoreapp3.1
$ ./Uno.UI.RemoteControl.Host --httpPort=9876

(The fully-working script lives in my dotfiles here)

The bottom-right pane is running the Skia.Gtk target, configured to use the Remote Control Host running on port 9876 (or whatever port number you specify):

$ dotnet build -property:UnoRemoteControlPort=9876 && \
    dotnet run --no-build

Note that dotnet run does not support specifying additional project properties as arguments, but dotnet build does, so this command separates out the build and run steps. The alternative would be to open the project file and add the UnoRemoteControlPort property manually, which works too, but would need to be updated any time you change the port or want to disable the hot reload feature.


Having seen how useful hot reload can be for desktop development, I started to wonder how to accomplish something similar using plain GTK. It supports an XML-based UI definition language too, so it shouldn’t be too difficult to support a similar feature.

It requires a little bit of extra code, but I can confirm that it is indeed possible!

This version of hot reload does not use a separate server; rather, the application itself monitors the UI file for changes, and when a change is detected, it re-invokes the rendering function. My example is in C, but it’s pretty short, and the same technique can be easily transferred to any language with GTK bindings:

// main.c

#include <gtk/gtk.h>
#include "hot-reload.c"

static void on_load_main_window(
	GtkBuilder *builder,
	GtkApplicationWindow *window
) {
	GObject *content_box = gtk_builder_get_object(builder, "content");
	gtk_window_set_child(GTK_WINDOW(window), GTK_WIDGET(content_box));

static void on_unload_main_window(
	GtkBuilder *builder,
	GtkApplicationWindow *window
) {
	g_info("Unloading main window");
	// Nothing to do explicitly (I think) if just replacing the window's child

static void activate(GtkApplication *app, gpointer user_data) {
	GtkWidget *window = gtk_application_window_new(app);
	gtk_widget_set_size_request(window, 400, 300);

The hot_reload() method creates a builder for main.ui, invokes the specified load callback, and then constructs a GFileMonitor that listens for changes to it. When a change is detected, the unload callback is invoked (if provided), and then the builder is re-created, and the load callback re-invoked.

The full example can be seen here, though I am not a professional in C, so don’t be too surprised if I missed a memory leak somewhere. It is intended to serve primarily as a proof-of-concept, rather than a full-fledged implementation.