GSoC part 8: the LED stack page
Jul 14, 2017
10 minute read

GSoC logo horizontal

Originally the plan for this week was to start working on the button mappings, however together with my mentor I decided that it’s better to do the LEDs first. This is because I was sure I could finish this in a few days, and button mappings is definitely going to take much longer than that. So, this week I’ll run you through the implementation of the LED stack page, and the coming weeks I’ll be working on button mappings, profile support and a proper welcome screen, in that order.

The LED stack page

A picture is worth more than a thousand words. A video, however, is priceless:

As you can see, I deviated from the mockup. The mockup uses multiple Gtk.ToggleButtons that are pressed (active) to highlight the current mode. One thing that wasn’t nice here is that to change a current mode (e.g. to change the solid mode’s color), you’d have to click an already active button. This isn’t obvious and so we iterated for a bit to find a better approach, until we landed on what you see in the video. Note: the off page still needs an image to indicate its empty state. If this is something you can help out with, please get in touch!

The pull request notes the issues I stumbled upon while implementing the LED stack page and its dialog, and the iterative design bit.

Firstly, changed values (even committed to the device) would not be remembered in between dialog invocations. At first, I thought this was due to the fact that the bindings only checked their DBus object’s properties upon instantiation, and not whenever they are retrieved (so a new setting would simply remain invisible). However, even after making sure that the latest value is always queried directly, things still didn’t work. We then found that ratbagd exposed the properties as constants, but this still didn’t fix the issue. As it turned out, Gio.DBusProxy wasn’t updating its cached properties because ratbagd wasn’t emitting property changed signals for all of its properties. My mentor fixed this overnight (for me at least; globally distributed software engineering at its finest!), only for us to run into the next issue… 😅

As you can see in the video, the button label should change to reflect the new LED mode after the user accepts the changes in the dialog. For some reason this wasn’t happening and the old label would be set, even though the dialog returned the correct mode and the method that converts a mode to its string representation was correct. Indeed, the correct mode was also being applied to the device. At first I thought this was a pass by reference versus a pass by value bug, but Python should support passing an object around and changing its properties just fine. A quick test confirmed my thinking (yay, I’m not going crazy just yet!); this does exactly what you’d expect it to:

class Foo(object):
  def __init__(self, x):
    self._x = x

  @property
  def x(self):
    return self._x

  @x.setter
  def x(self, xx):
    self._x = xx

def modify(foo):
  foo.x = 10

foo = Foo(5)
print(foo.x)
modify(foo)
print(foo.x)

With some more thinking, even a pass by value versus a pass by reference would be weird, since the changes are being applied correctly and the callback that is called when an LED’s mode changes was being called on the correct LED object. I approached my mentor with this issue, who, with his years of experience (😉), was quick to smell a race condition in DBus.

  1. We set the LED mode on DBus though the SetMode(mode) method on the LED object;
  2. We query the DBus LED mode, which returns the old mode because that’s what the Gio.DBusProxy has;
  3. Time passes, and we return to GLib’s main loop which monitors DBus;
  4. DBus’ property changed signal arrives;
  5. If we query the DBus LED mode now, it has the correct value.

Because there is no time between steps 1 and 2, the signal hasn’t yet arrived and the cached property isn’t updated. The solution thus is to explicitly and immediately update the cached DBus properties when a setter method is called. However, you only ever want to update a cached property if its setter method actually applied the changes, so we needed to refactor _RatbagdDBus._dbus_call. There were a bunch of options here (e.g. catching any DBus exceptions and simply returning different error codes), but in the end I went for raising different kinds of exceptions signaling different ratbag error codes returned from the DBus method. I went this way because it 1) automatically prevents the cached property from being updated when an exception is raised, 2) is automatically propagated up the call chain to the UI (or any other user) that can then decide what to do, and 3) it is the most Pythonic.

Finally, Piper now checks if the device supports resolutions and LEDs before it shows those pages. Ratbag exposes device capabilities for precisely this reason, but up until now I had simply forgotten about them. To make these capabilities easier to work with (my device would support CAP_SWITCHABLE_RESOLUTION but not CAP_RESOLUTION), my mentor made sure that “parent capabilities” such as CAP_RESOLUTION are set before any child capabilities such as CAP_SWITCHABLE_RESOLUTION are.

Setting a default resolution

A leftover from last week’s resolution stack page is that there is no way to set the default resolution yet. This week I attempted to fix that, through this PR. There may at any given time be only exactly one default resolution, and hence a Gtk.RadioButton initially seems like the best widget to use here given that it itself enforces this behaviour so we don’t have to do that manually. However, it didn’t look very nice and its intention also isn’t clear from a usability point of view.

To make it look prettier and to improve the usability, my mentor suggested I move the widget into the revealer that appears when a resolution is clicked from the list. This also gives us more space so we can add a label describing what the widget does. Initially I went with a Gtk.Switch but later changed this to a Gtk.CheckButton.

This does mean that we now need to manually enforce that there is only exactly one default resolution at any given time, but fortunately this isn’t that difficult:

@GtkTemplate.Callback
def _on_default_toggled(self, check):
    # The user toggled us; set ourselves as the default.
    if check.get_active():
        self._resolution.set_default()

def _on_default_resolution_changed(self, resolution, default_index):
    if default_index == self._resolution.index:
        # It's us; set ourselves insensitive so that the user cannot have
        # zero default resolutions.
        self.default_check.set_sensitive(False)
    elif self.check.get_active():
        # We were the old default; unset ourselves and make ourselves
        # sensitive again.
        self.default_check.set_sensitive(True)
        self.default_check.set_active(False)

All of this would work nicely, if only the DefaultResolutionChanged signal would be received by the RatbagdResolution objects. As it turns out, this isn’t happening because there is one such object for each resolution, with object paths /org/freedesktop/ratbag1/resolution/event13/p0/r0 (where p0 and r0 encode the profile and resolution to encode for different profiles and resolutions, and event13 is the mouse under configuration) with interface org.freedesktop.ratbag1.Resolution, while the DefaultResolutionChanged signal is emitted by ratbagd on /org/freedesktop/ratbag1 with the same interface. Gio.DBusProxy says the following for its g-signal signal: Emitted when a signal from the remote object and interface that proxy is for, has been received. So because the signal it receives isn’t for its object path (.../resolution/event13/p0/r0) it doesn’t emit its g-signal.

The fix initially appears simple: just emit the signal on the resolution object’s path, so that it will receive it. That’s exactly what I did, but as you can read there the next issue is that the signal is now only emitted on the resolution that has just become the default. At the very least it should also be emitted on the signal that was the default, so that it knows it isn’t anymore.

There are two ways to fix this issue:

  1. Move the DefaultResolutionChanged (and ActiveResolutionChanged) signal up into the profile, or
  2. Have a boolean property of IsDefault in the resolution objects. This way we get to use DBus’ PropertyChanged signals as they happen anyway, without the need for extra signals on our behalf.

The latter would mean that each class in Piper that has a Ratbagd* instance (let’s say a RatbagdResolution in this example) needs to connect to that instance’s notify::is-default signal that is emitted whenever this property changes (automatically, by GObject). This is straightforward, but I’m not sure how well it will work with profile switches.

The former would mean that each class in Piper that has a Ratbagd* instance needs a reference to a profile, so that it can connect to the profile’s default-resolution-changed signal (again, just taking the RatbagdResolution example here) and do the work from that callback. This will likely work better with profile switches, albeit being a little less straightforward.

I’m not yet sure which of these will work best once I will implement profile switching: there won’t be just one profile to work with but all of them, and the active one can change at any given time. In this case, it is likely that every class in Piper needs a reference to a profile anyway, which will change depending on the active-profile-changed signal emitted from RatbagdDevice (currently RatbagdProfile, but this will have to be changed to prevent the same issue we’re having now with RatbagdResolution) to grab, for example, the new profile’s resolution to update.

My suggestion is thus to postpone this issue until I get around to implementing profile support in a few weeks, so that we only have to change things once.

Smaller changes

Last week I asked the question does every resolution have its own report rate?. The concensus then was to implement report rates per profile, even though this issue was opened not much later. It is now definitive that supporting per-resolution report rates isn’t required, so I made the change to make it per-profile. It is however not much work at all to make it work per-resolution, so for devices that support this we might just want to do it after all:

diff --git a/piper/resolutionspage.py b/piper/resolutionspage.py
index 9c7faa6..c0b580f 100644
--- a/piper/resolutionspage.py
+++ b/piper/resolutionspage.py
@@ -75,11 +75,11 @@ class ResolutionsPage(Gtk.Box):

     def _on_report_rate_toggled(self, button, rate):
         profile = self._device.active_profile
-        # TODO: currently no devices expose CAP_INDIVIDUAL_REPORT_RATE, but if
-        # so then we should check for this here and set it only on the relevant
-        # resolution.
-        for resolution in profile.resolutions:
-            resolution.report_rate = rate
+        if RatbagdResolution.CAP_INDIVIDUAL_REPORT_RATE in profile.active_resolution:
+            profile.active_resolution.report_rate = rate
+        else:
+            for resolution in profile.resolutions:
+                resolution.report_rate = rate

     @GtkTemplate.Callback
     def _on_row_activated(self, listbox, row):

In Python methods can return more than one value. In some cases, you are only interested in some of them, in which case it is common to use an anonymous variable which is often named _. Piper did this as well, but Piper is also (planning on being) internationalized. In internationalized applications, it is common to alias the gettext() function to _, so that strings can be marked for translation using the _(<some string>) syntax. Using an underscore for both gettext() and anonymous variables can be confusing and lead to interference, so Piper now uses a double underscore for anonymous variables.

Last week’s resolutions page contained a small bug that would mess up the toggling of rows in the list: if toggling the same row twice and then toggling another row, both the first and the second row would be toggled. The fix was rather straightforward and required resetting last_activated_row to None in case the same row is toggled.

A widget’s purpose might not always be immediately clear. In such cases, a tooltip helps to understand what a widget does. As such, I have added a tooltip to each control widget currently found in Piper.

Finally, we now have toned down device SVGs for the Logitech G500 and G500s and soon for the Logitech G303. Thanks!

This blog post is part of a series. You can read the next part about the button stack page here or the previous part about the resolutions stack page here.