Bug 105472

git-bisect shows that the culprit was e43bd4ae3a75f72b45c6bab6168bd859ab4a01b3 (plus e8dffbd73a1b3c17716f972f210e420de94028c2, obviously).

I can't help but notice that a simple on/off switch would probably have been a heck of a lot simpler: less code, higher reliability, and no regressions for anyone, instead of this push-pull between people with good hardware who don't want hysteresis, and those with crappy hardware who do. We'd probably have had this issue solved long ago, rather than playing whack-a-mole with it into the foreseeable future.

And now I promise that will be my last comment on the subject in this bug report. :)

Except not really that much simpler. On/off isn't enough, you'll need to provide some hysteresis range. Oh, now you need it per-axis because some devices jitter on the y axis only. Oh, and you cannot ever switch away from the hysteresis to something better because you promised that the configuration exists. You cannot change the hysteresis algorithm, because config settings are written in stone. Want to change from physical units to device units? Oops, cannot do that without adding another option. And making sure you are compatible between the two.

You also have to expose the config option, write a GUI for every desktop. Deal with the corner-cases of toggling the config option at the 'wrong' time.

Then you have to teach users that that option exists. Where it is, in every desktop and how to set it. In 3 years time, the ubuntu forums are full of posts suggesting random config options that may not even exist anymore. 

And AFAICT neither windows or macos have that setting, so they're smart enough to figure this out. Oh, and having the toggle to turn it off also stops us from thinking about the correct solution.

An on/off switch makes things easy for the moment, for that one user that wants it. It makes life a lot more difficult for maintainers. So please, let's stop the talk of "just add a config option" because yes, for you it would've been easier (right now). For me and a lot of others, it's just a long-term waste of time.

I understand. It's just frustrating to still be wrestling with this issue a year and a half later, after the herculean effort you've put in to make it work. Hysteresis is just so frustrating with my hardware that it makes me want to chuck the computer out a window.

Anyway, I'm done grousing and am ready to help! :)

> if I continuously make small circles with the cursor

Is this a real use-case? You can trick the code (it's a result of how simple it is) but I haven't been able to trigger this during normal interaction.

> Is this a real use-case?

No, it's just the best way to quickly tell if hysteresis is on or off.

Your device should have the hysteresis disabled by default. So by doing that you're tricking the code into thinking you have pointer jitter and the wobble detection code *enables* the hysteresis. You're defeating yourself here :)

Let me know if the hysteresis enables during a real-world use-case, because that's something worth investigating.

Um, how do I turn it off then? It turned on quite before I did the small circles routine; I only did that to illustrate the issue. I really don't like how easy it is to accidentally turn this on.

There's no toggle to turn it off right now. Best solution may be to keep monitoring the touchpad and build up heuristics to make the guess whether it wobbles or not more reliable. E.g. restore the old "not wobbling" check together with the new "this is wobbling" check and then pitch the two against each other.

It turned off again and hasn't turned back on, so maybe I was overthinking things. I'll open a new bug if I find that it's turned on inappropriately during normal use.