mirror of
https://github.com/bryanthaboi/gen1recomp.git
synced 2026-08-12 00:10:56 +02:00
bryanthaboi
72 lines
2.9 KiB
Lua
72 lines
2.9 KiB
Lua
-- #781: Linux launcher mouse-dead behind the pad cursor. Reproduces the
|
|
-- X11 multi-monitor failure mode (polled love.mouse.getPosition frozen on
|
|
-- desktop-virtual coords, so the motion yield in _updatePadCursor never
|
|
-- fires) and asserts a host-forwarded mousepressed reclaims the pointer.
|
|
-- Self-contained: `luajit tests/rom_importer_cursor_bug781_test.lua`.
|
|
-- Should eventually merge into tests/rom_importer_cursor_test.lua (dofile'd
|
|
-- by tests/run_tests.lua).
|
|
package.path = "./?.lua;./?/init.lua;" .. package.path
|
|
if not _G.love then _G.love = require("tests.love_stub") end
|
|
|
|
local S = require("tests.harness").suite("rom importer pad cursor #781")
|
|
local eq = S.eq
|
|
|
|
local RomImporter = require("src.import.RomImporter")
|
|
|
|
-- Bare importer with just the pad-cursor state new() would build; isNX
|
|
-- false keeps _updatePadCursor on the desktop path (polled motion yield),
|
|
-- _flex nil keeps the right-stick branch out of LauncherView.
|
|
local function makeImporter()
|
|
return setmetatable({
|
|
android = false,
|
|
isNX = false,
|
|
_flex = nil,
|
|
_padCursor = { x = 320, y = 260 },
|
|
_padCursorActive = false,
|
|
_padAxis = { leftx = 0, lefty = 0, righty = 0 },
|
|
_padDir = {},
|
|
_padInited = true,
|
|
}, RomImporter)
|
|
end
|
|
|
|
-- Failure mode: SDL's polled mouse state stuck on coordinates outside the
|
|
-- window (primary display away from desktop 0,0). Successive samples are
|
|
-- identical, so the motion yield sees zero delta and never releases the
|
|
-- pad cursor no matter how much the real mouse moves.
|
|
local ri = makeImporter()
|
|
love.mouse.getPosition = function() return 2960, 4130 end
|
|
ri._padCursorActive = true
|
|
ri:_updatePadCursor(1 / 60) -- seeds _lastMouseX/_lastMouseY
|
|
ri:_updatePadCursor(1 / 60)
|
|
ri:_updatePadCursor(1 / 60)
|
|
eq(ri._padCursorActive, true,
|
|
"frozen polled coords starve the motion yield (the #781 trap)")
|
|
|
|
-- The fix: the host-forwarded real press must win the pointer back, same
|
|
-- contract as PadCursor.yieldToPointer in the overlay hosts. This is the
|
|
-- half that un-gates LauncherView.update's click minting.
|
|
ri:mousepressed(10, 10, 1)
|
|
eq(ri._padCursorActive, false,
|
|
"mousepressed reclaims the pointer even when the yield is starved (#781)")
|
|
|
|
-- A reclaimed pointer must stay reclaimed: the next pad-cursor tick with
|
|
-- still-frozen polled coords may not re-arm it by itself.
|
|
ri:_updatePadCursor(1 / 60)
|
|
eq(ri._padCursorActive, false,
|
|
"an idle pad tick does not re-steal the pointer after reclaim")
|
|
|
|
-- Regression guard for the healthy desktop path: when polled coords do
|
|
-- move (window-relative, single monitor), the existing motion yield still
|
|
-- releases the pad cursor without needing a click.
|
|
local ri2 = makeImporter()
|
|
local px = 100
|
|
love.mouse.getPosition = function() return px, 100 end
|
|
ri2._padCursorActive = true
|
|
ri2:_updatePadCursor(1 / 60)
|
|
px = 140
|
|
ri2:_updatePadCursor(1 / 60)
|
|
eq(ri2._padCursorActive, false,
|
|
"real mouse motion still yields the pad cursor on sane polled coords")
|
|
|
|
S.finish()
|