Repository navigation
Conversation
…r long it is A swipe up went home only when it stayed under 120 px. Past that it opened recent apps, and an ordinary swipe is longer than that, so going home took a deliberately short flick and every natural swipe landed in recent apps. Now a swipe goes home however far the finger travels, and recent apps open only when the finger stops and rests for 350 ms while still down, as on iOS and Android. The rest is measured against where the finger was when the wait began, with 8 px of slack, so the small shifts of a resting finger don't start the wait again and a slow, steady swipe is not taken for a rest. A swipe taken back down under 24 px before it rests does nothing, as before. test/pill-touch.qml gains four cases under touch: a 360 px swipe goes home, a swipe held still opens recent apps once and doesn't also go home, a slow swipe is no hold, and a swipe taken back does nothing. All four fail against 0.2.133's code.
…omes up is put away A swipe home from an application with the keyboard up left the keyboard over the home screen when the application kept its text field. A GPUI application does not give its text input up when its window loses the keyboard focus, and on a Pixel 9a (Hyprland 0.57) the input method got no deactivate when an empty workspace took the focus, so the keyboard had no reason to go. GTK and the browser disable their text input themselves, which is why it did not show with them. Until now the plugin put away only a keyboard it had shown by hand. The keyboard plugin now hides whatever keyboard is up at the moment the home screen comes up (KeyboardModel.hideAsHomeComesUp). One that comes up later on the home screen, for the drawer's search field, is not touched: the home screen was up already. Tested: test/keyboard-model.js (a keyboard up for a text field is put away as the home screen comes up; one that comes up on the home screen stays; the moment is the home screen coming up, not staying there or leaving) and test/all. On a Pixel 9a, with the plugin's two files put in place by hand and the shell restarted: with the keyboard up over a GPUI application, focusing an empty workspace hid it (the shell's log has "home screen up, hiding the keyboard", and sm.puri.OSK0's Visible read false after). That keyboard was Tactus, which exports squeekboard's interface; with squeekboard itself it has not been tried.
|
Thank you. Taking the 120 px rule out is right, and you found why it was there: on main the hold's timer starts again on every pixel the finger moves, so under a real finger it almost never fired and "swipe far" was the only dependable way to recent apps. Your 8 px of slack is what makes the hold usable, and with that the long swipe can go home as it does on every other phone. I read the four new touch cases against main's handler and each would fail there, as you say. I also like the second commit: whatever keyboard is up at the moment the home screen comes up goes away, and one that comes up afterwards for the drawer's search stays. The model function is small and the test says exactly that. This changes a gesture people make dozens of times a day, on every phone, so I'll run it on my Pixel 7 Pro before it merges, as the installed package:
Two things I'd like you to look at meanwhile, since you use it daily. First, "the home screen comes up" is "the focused workspace has no window". Until now a wrong moment there could only hide a keyboard the plugin had shown itself; now it hides any keyboard. Have you seen the keyboard drop while an app was still in front, for instance when an app's first window closes as its second opens, or when you change apps through recent apps with the keyboard up? Second, your phone tests were on an installed On the version: #886, #888, #890 and #204 also say 0.2.134. I expect #886 to go first, so this one will need 0.2.135 on the branch before I merge it; I'll confirm the number when the 7 Pro run is done. |
|
Answers from daily use on my palmtop, which has run this change since this morning (the launcher's gesture and the keyboard plugin's
The |
Over an app, a swipe up on the home pill went home only if it stayed under 120 px. Anything longer opened recent apps, and an ordinary swipe is longer than that, so every natural swipe up landed in recent apps and going home took a deliberately short flick. iOS and Android both go home on a swipe and keep recent apps for a swipe that stops and rests.
This takes the 120 px rule out. A swipe goes home however far the finger travels, and recent apps open when the finger rests for 350 ms while it is still down. The rest is measured with 8 px of slack from where the wait began, so a resting finger's small shifts don't start the wait again and a slow, steady swipe isn't taken for a rest. A swipe taken back down before it rests still does nothing. The dock's swipe on the home screen is unchanged.
A second commit goes with it, because a swipe home is now the ordinary way out of an application: any keyboard still up when the home screen comes up is put away. With the keyboard up over a GPUI application, a swipe home left the keyboard over the home screen. That application keeps its text input when its window loses the keyboard focus, and on my 9a (Hyprland 0.57) the input method got no deactivate when an empty workspace took the focus, so the keyboard had no reason to go. GTK and the browser disable their text input themselves, which is why it doesn't show with them. Until now the keyboard plugin put away only a keyboard it had shown by hand; it now hides whatever keyboard is up at the moment the home screen comes up, and leaves alone one that comes up later, for the drawer's search field.
Tested:
test/pill-touch.qmlhas four new cases under touch: a 360 px swipe goes home, a swipe held still opens recent apps once and doesn't also go home, a slow swipe is no hold, a swipe taken back does nothing. All four fail on 0.2.133's code.test/allpasses, andtools/pr-check origin/main.7.3.0-rc6-omamobile-tegu-s2mpg15-m1-g991042e2eff3-dirty), with the same change put into the installedLauncher.qmland touches put into the touchscreen's own event stream: a 500 px swipe in 200 ms went home (the log line isswipe -500with norecents), and a 200 px swipe held for 0.7 s, the finger shifting by a pixel or two, opened recent apps while the finger was still down (swipe -202 recents). I've used it that way by hand since.test/keyboard-model.js(a keyboard up for a text field is put away as the home screen comes up; one that comes up on the home screen stays; the moment is the home screen coming up, not staying there or leaving). On my 9a, with the plugin's two files put in place by hand and the shell restarted: with the keyboard up over a GPUI application, focusing an empty workspace hid it (the shell's log has "home screen up, hiding the keyboard", andsm.puri.OSK0'sVisibleread false after).tests/allpasses on the branch.Not tested: the keyboard change with squeekboard itself (the keyboard on my 9a exports squeekboard's interface, but is not squeekboard); the package installed as a package on a phone (the change ran as an edit of the installed file, on a phone whose home strip I have restyled; the gesture code is this one), and no phone but the 9a.
tests/packagesfails in omamobile-omarchy on my laptop with and without this change (qmlformat on two of the installed Omarchy's files), nothing else.It bumps omamobile-shell to 0.2.134, as #886 does; whichever lands second gets a rebase from me.