DioxusLabs
Save logical position when restoring on macos
On macOS, it behaves strange with multiple monitor physical location. That is to say
Consider we have 2 monitor main and extra which physical size is 2560x1440 (scale: 2.0) and 1920x1080 (scale: 1.0)
+-----------+---------+
| .<-A | .<-B |
| main | extra |
| 2560x1440 |1920x1080|
| |---------+
+-----------+
The physical position of A is (1200, 100), and the logical position of it would be (600, 50) due to scale factor.
The code before this PR would save (1200, 100) to "window-state.json" and everything just works.
But for point B, which "relative" physical position is (100, 100) (offset from left-top of "extra" monitor),
In this case, winit would tell the physical position is (1380, 100), which is the same as logical position (1380, 100).
If we save (1380, 100) to state file, and when restore, the winit would think we are at main screen and position to (640, 50) in logical position.
| Point name | p1 = window.outer_position() |
p2 = p1.to_logical(window.scale_factor()) |
restored p1 (in logical) | restore p2 |
|---|---|---|---|---|
| A | PhysicalPosition(1200, 100) | LogicalPosition (600, 50) | LogicalPosition (600, 50) | LogicalPosition (600, 50) |
| B | PhysicalPosition(1380, 100) | LogicalPosition (1380, 100) | LogicalPosition (640, 50) | LogicalPosition (1380, 100) |
TL;DR, in macOS the position system is consist in "logical" coordinates.
Edit:
This seems only macOS related, so I restrict patch only on cfg!(target_os = "macos")
There're several related issues you could reference
- https://github.com/tauri-apps/tao/issues/816
- https://github.com/rust-windowing/winit/issues/2645
- https://github.com/tauri-apps/plugins-workspace/issues/2620
I found let adjustment = 56 failed on extra monitor.
Update:
It should be LogicalSize(0, 28) empirical.
(though I cannot find any documentation, the original code introduced in https://github.com/DioxusLabs/dioxus/commit/e923c6462c497a55388a62b93373fba972847594, could @jkelleyrtp kindly explain why it used to be 56 here?)
Tested on my macOS (15.4.1) and windows (11 24H2)
Thanks for the PR!
Should we implement LogicalPosition for all platforms, not just macos? Does it make sense to maintain both?
I believe the 56 offset is meant to handle the thickness of the topbar, otherwise it marches down on a reload.
Also CI not passing is blocking this PR, might be worth merging main to see if the failures go away.
On macos, there's 1-1 mapping from points on screen to points in Logical Position, (and there's 2 points on screen have the same expression in Physical Position, like (1380, 100)) While on windows, there's 1-1 mapping from points on screen to points in Physical Position.
I once tried use logical position on windows but it failed to restore to the same location if 2 monitors have different scale. So I revert it in https://github.com/DioxusLabs/dioxus/pull/4061/commits/0fbd886a950cc16a6d53cfa73e0d14486e0f890a
For CI failure, I think there's some network issue, can re rerun it?