GSoC part 13: I solved global warming!
Aug 18, 2017
9 minute read

GSoC logo horizontal

This week I have been solving more issues to make sure that Piper offers a pleasent user experience, doesn’t crash and runs smoothly. My mentor and cvuchener from the libratbag project have been testing Piper the last week, and together with a handful of users attracted to Piper they have opened a bunch of issues for me to solve. Let’s run through the most visible ones!

Solving global warming

Probably the most fun issue to resolve this week was the one reported by my mentor: Piper contributes to global warming (you won’t believe what happened next!). The issue here was that when the MouseMap is partially or fully obscured by another window, Piper’s CPU usage would touch the 100% and hence, contribute to global warming. Since global warming is bad (and yes, it is real 😉), obviously I had to fix it.

The report already contained a hint as to what was happening: the MouseMap’s do_draw method was being called a lot (with some imprecise measurements, I got it to roughly every 30 milliseconds). Through some more debugging I found that the issue went away if we didn’t retrieve the theme’s color for links from the MouseMap’s StyleContext. Gtk’s baedert on gtk+ linked me to this old bug report that explains what’s going on. What follows here is my summary of the linked blog post, that goes into more detail than the bug report:

Since Gtk 3.18, passing a state other than the current state to Gtk.StyleContext::get_color isn’t recommended and can, apparently, lead to hogging the CPU due to triggering a loop: requesting the style context’s color triggers a pixel cache refresh, which triggers do_draw, which requests the style context’s color, et cetera.

The fix, then, is straightforward: save and restore the style context around requesting a color, so that it won’t be invalidated and won’t trigger a pixel cache refresh. Indeed, this is the suggested (temporary) workaround given in the linked blog post; the given permanent solution is to use the gtk_render_* API. However, this doesn’t apply for the MouseMap and as such, I used the workaround to do my share of saving the planet.

Preventing the user from shooting themselves in the foot

During GUADEC my mentor opened the issue that we shouldn’t allow a user to unset the left mouse button, because if they do without mapping another button to a left mouse click they won’t have a left mouse click anymore and that’s… well, troublesome to say the least. The only question was how to best do this: randomly mapping another button to a left mouse click was undesirable (for obvious reasons) but we also shouldn’t outright disallow remapping the (physical) left mouse button as there are perfectly valid scenarios to do so. Think, for example, of left handed users (yours truly) that swap the left- and right mouse buttons to use the mouse with their left hand (not yours truly).

My first attempt then was to pop up a dialog that asks the user if they know what they’re doing when we detect there isn’t another button mapped to a left mouse click when they are attempting to remap the only/last button that is mapped to a left mouse click (savvy?). I suppressed the voice in my head that says dialogs are invasive and should be avoided, because I thought this was a good compromise between protecting the user while still giving them all the options. (As a sidenote, an undo button wouldn’t work either because when the last left mouse click has been unmapped, there is no left mouse click to press undo with anymore).

I quickly implemented a proof of concept, but my mentor was stronger than I was and didn’t give in to the dialog. Instead, he said we should for now only allow to swap the left- and right mouse buttons; we’ll hear it from the users when they want more options. After two design iterations, this is what I’ve settled on:

The relevant pull requests are #117 for the initial implementation and #143 (unmerged) for the UI rework.

Another way to shoot yourself in the foot is by assigning a button in one profile to cycle through profiles but not in another: eventually, you’ll get stuck in that profile with no way out but to open Piper. In the heat of battle where either it’s you or them, obviously you don’t want this to happen. This is exactly why my mentor reported this issue, again over GUADEC, and, as of last week, this won’t happen and Piper is just that bit smarter!

Preventing Piper from crashing

Libratbag contributor cvuchener reported a bunch of issues last week, in all of which Piper was crashing.

First of all, Piper was crashing when opening devices with weird profiles:

  1. A profile containing an unsupported special button mapping crashed Piper, and
  2. A device with profiles that did not have an active resolution set.

With Piper being a graphical user interface to libratbag it naturally cannot support anything that libratbag doesn’t, but it also shouldn’t crash when it does see something that isn’t supported. In this case, the solution is to recognize that there are unsupported actions and to simply add the RatbagdButton.ACTION_SPECIAL_UNKNOWN entry to the map in the bindings that maps special button mappings to their human readable representation.

While testing this, my device suddenly got a special button mapping with value 4294967295 (you’ll have to take my word for this value not having any significance within libratbag). As it turns out, libratbag (or ratbagd) doesn’t perform a range check and can thus let arbitrary special button mappings go through instead of setting or returning RatbagdButton.ACTION_SPECIAL_UNKNOWN for all unknown values. This can be solved in Piper by catching any KeyErrors when looking up a mapping’s human readable representation, but it’s better to fix this in libratbag itself; which is why I opened this issue.

As for the second crash mentioned above, apparently there are some misconfigured libratbag drivers that do not set an active resolution. In this case, Piper now prints a message informing the user of a buggy driver and simply returns the first resolution (or profile) it finds (#122).

Cvuchener also found that the apply button was switching state when going back to and from the welcome screen. This was simply a case of me trying to be too smart for my own good, as you can see in the one line fix.

I also forgot to destroy old profiles when reusing the MousePerspective which led to duplicate profile entries that crashed Piper when clicked.

Finally, cvuchener ran into timeouts when committing changes to his device, which has been a reoccurring issue for myself as well. My mentor and I had discussed async commits before, but until libratbag supports this there isn’t much that Piper can do. Deciding on a proper timeout is difficult, as we

  1. do not know what happens when a user changes settings while old changes are being committed (so we should strive to make it as small as possible?), and
  2. don’t know what value will be long enough not to cause timeouts. This depends both on the device (including whether it’s wireless or not, as for example my G403 wired does not lead to timeouts but wireless it does) and the box running Piper.

We assume that DBus will cache any changes for us, and indeed it seems to work just fine, so as a stop-gap until we get async commits the timeout is now 2 seconds instead of half a second.

This week we’ve seen a bunch of users trying to run Piper. This is exciting because it means we’re doing work that people need! It is also good because they run into issues that need fixing before we tag a release ☺

One such case is a user that tried to run Piper with a too old Gtk version. This resulted in us adding a check to Piper for the required Gtk version. Next, when that user upgraded his Gtk version, he made libratbag recognize his device but forgot to supply libratbag with a device SVG. Piper should have displayed the error perspective in such a scenario, but it turned out that we cannot raise exceptions from GObject properties. After making this a real setter method, the error perspective is now presented as it should.

Work-in-progress

I’m still working on a bunch of other issues that I can hopefully get merged before the 21st of August, which marks the end of my Google Summer of Code. I intend to keep working on Piper after the summer, but these are features I would like to get in before we tag a release.

First, jimmac requested a search field for the button mapping dialog. Today I opened the pull request that adds this and makes the dialog in general look just that bit nicer (compared with what I showed part 10):

Another GUADEC issue was to highlight the active resolution in the resolutions page. This was a straightforward change, after having fixing ratbagd (#284). Here you can see it in action:

As you can see, this activates the clicked resolution and thereby also fixes that issue (but ignores issue #72). This approach should also work towards being able to display and set the default resolution, a pull request that has been WIP since July 11.

Last, but definitely not least, is that Piper can now be translated into your native (or not!) language! Here you can see Piper in Dutch:

As you can see, not everything is translated yet. There is something weird going on with all strings that are translated and retrieved from a map; I have yet to figure this one out. We also cannot yet run with local translations: for now you’ll have to install Piper in order to test your translations. I also intend to add more context for translators so it is easier to see how to translate some strings, and I intend to get this merged only after all the UI work is done so we are sure to catch all strings.

Barring any newly discovered crashes, there are only three more issues to solve before my deadline: a crash resulting from libratbag returning its error code 1001, correctly restoring the macro preview label and fixing the initial window size.

This blog post is part of a series. You can read the next part about the final changes here or the previous part about the finishing touches here.