Release notes

BUGFIX

Panel switching keeps your window on its monitor

After you drag Usage's panel (a small floating pop-up) to a second monitor, switching its style or theme will finally leave it there. Windows had only handed the app the primary display's usable area, so each reposition used the wrong ruler and shoved the panel back onto the primary monitor. It now identifies the usable area for the display containing the panel's current or remembered position. On a single monitor, it also retains the previous style's measured height while the new one is measured, instead of briefly using a near-fullscreen placeholder that pushes your window upward. Like movers checking which building you live in before returning furniture, rather than always dumping it in the lobby.

Original changelog

Original changelogPanel switching keeps your window on its monitor
**The Windows panel no longer snaps back to the primary monitor on every panel switch**: dragging the popover onto a second display and then switching panel style (or theme) forced it right back onto the primary monitor. Windows' `SPI_GETWORKAREA` API only ever reports the *primary* monitor's work area, and the window was clamped against that single rectangle on every reposition regardless of which monitor it actually lived on — so a window dragged onto a secondary display got clamped straight back onto the primary one the next time the panel reloaded. Placement now looks up the work area of whichever monitor the window's current (or saved) position falls on via `MonitorFromPoint`/`GetMonitorInfoW`, so a window on a secondary display stays there across panel switches. A related issue on a single monitor is fixed alongside it: switching panels reset the remembered content height to `None`, so the window was briefly clamped against each panel's near-fullscreen placeholder height (`PANEL_HEIGHTS`) before its real height was measured, which could shove a dragged window toward the top of the screen; the previous panel's measured height is now kept as the transitional estimate instead.