fix regressions with freemove

This commit is contained in:
DramaticShape
2026-08-08 10:42:28 -04:00
parent 41f1f07342
commit dfe2f1ae1a
5 changed files with 299 additions and 10 deletions
+95 -10
View File
@@ -67,7 +67,13 @@ return function(game)
-- this driver has to be able to see (button edges read on the render
-- clock instead of the logic step) lives between love.keypressed and
-- whoever polls the edge, so a driver that writes the queue itself
-- jumps straight over it. "b" is the default binding for the B button.
-- jumps straight over it.
--
-- The B BUTTON's keyboard binding is "x" (or backspace) -- Input's
-- DEFAULT_BINDINGS. The `b = "b"` next to it is the GAMEPAD table, so a
-- driver that presses the "b" KEY presses nothing at all and reports a
-- dead button whatever the code does.
local B_KEY = "x"
local function key(name)
love.keypressed(name, name, false)
U.wait(2)
@@ -90,6 +96,20 @@ return function(game)
U.wait(30)
end
-- Tap the fight all the way off the stack before the next case starts.
-- Cases that merely tapped A a fixed number of times left the previous
-- battle (and its session) alive whenever an outcome ran long, and the
-- next case then measured the leftover -- which reads as that case
-- failing, at a spot nowhere near the cause.
local function closeOut(battle)
for _ = 1, 400 do
if not CatchThrow.session() and game.stack:top() ~= battle then break end
U.tap(game, "a")
U.wait(4)
end
U.wait(40)
end
local function describe(tag, battle)
local s = CatchThrow.session()
U.log(("%s: session=%s empty=%s phase=%s battle=%s veil=%s hidePlayer=%s")
@@ -116,7 +136,7 @@ return function(game)
U.shot(game, DIR .. "/empty_1_aim.png")
-- and B is the way out: a Let's Go wild always escapes
key("b")
key(B_KEY)
for _ = 1, 120 do
U.wait(2)
if a.result then break end
@@ -126,8 +146,7 @@ return function(game)
-- the session and the held camera are swept by battle.ended, which is the
-- teardown BELOW this, not the moment `result` is written -- so the sweep
-- is only worth asserting once the battle has actually left the stack
for _ = 1, 40 do U.tap(game, "a"); U.wait(4) end
U.wait(60)
closeOut(a)
U.log(("A: after teardown -- session=%s veil=%s (want gone/down)")
:format(CatchThrow.session() and "still up" or "gone",
BattleScene.capture and "still up" or "down"))
@@ -139,6 +158,13 @@ return function(game)
U.log("== B: FULL with " .. ballCount() .. " ball")
local b = BattleState.newWild(game, "PIDGEY", 5)
b.onFinish = function() end
-- The outcome under test is what happens when the LAST ball MISSES, so
-- the roll must not be left to chance: a run where the Pidgey happened
-- to be caught used to cascade into every later case (the session lived
-- on into its epilogue, and C and D then measured that leftover instead
-- of their own). Forced failure, three shakes -- the roll is the
-- engine's business and is covered elsewhere.
b.catchAttempt = function() return false, 3 end
game.overworld:pushBattle(b)
toCapture(b)
describe("B", b)
@@ -186,15 +212,14 @@ return function(game)
(sb and ("still holding " .. tostring(sb.ballId)))
or "SESSION GONE"))
U.shot(game, DIR .. "/empty_2_ranout.png")
key("b")
key(B_KEY)
for _ = 1, 120 do
U.wait(2)
if b.result then break end
U.tap(game, "a")
end
U.log("B: after B -- result=" .. tostring(b.result) .. " (want run)")
for _ = 1, 40 do U.tap(game, "a"); U.wait(4) end
U.wait(60)
closeOut(b)
-- ------- C: the scripted tutorial, untouched
@@ -256,7 +281,68 @@ return function(game)
toCapture(d)
local sd = describe("D" .. speed, d)
if sd and not sd.empty then
key("b")
-- where does the key actually get to? Each stage of the chain, so a
-- dead B is attributed rather than guessed at
local top = game.stack and game.stack:top()
U.log(("D%dX probe: top=%s onKeyPressed=%s fullWild=%s declinable=%s")
:format(speed, tostring(top and top.screenId or "?"),
tostring(top and top.onKeyPressed ~= nil),
tostring(sd.fullWild), tostring(sd.declinable)))
-- ------- the counter-factual, measured rather than argued
--
-- How many times an edge would have been visible to a poll on the
-- RENDER clock -- where these reads used to live. Measured on UP
-- rather than B: the fix TAKES B out of the queue, so B never
-- reaches `pressed` any more and would read as a false zero. UP is
-- untouched by everything while the battle is parked, and the
-- question is about step-vs-frame ordering, not about which button.
local onFrame, onStep = 0, 0
local innerU = CatchThrow.update
CatchThrow.update = function(dt)
if game.input.wasPressed and game.input:wasPressed("up") then
onFrame = onFrame + 1
end
return innerU(dt)
end
local innerB0 = CatchThrow.buttons
CatchThrow.buttons = function(g)
local q = g and g.input and g.input.pressQueue
if q then
for i = 1, #q do if q[i] == "up" then onStep = onStep + 1 end end
end
return innerB0(g)
end
love.keypressed("up", "up", false)
U.wait(2)
love.keyreleased("up", "up")
U.wait(2)
CatchThrow.buttons = innerB0
CatchThrow.update = innerU
U.log(("D%dX counter-factual: one UP press -- the logic step saw it "
.. "%d time(s), the RENDER clock %d time(s)%s")
:format(speed, onStep, onFrame,
onFrame == 0 and " <-- a frame poll misses it entirely"
or ""))
local seen = 0
local innerB = CatchThrow.buttons
CatchThrow.buttons = function(g)
local q = g and g.input and g.input.pressQueue
if q and #q > 0 then
seen = seen + 1
U.log("D probe: buttons saw queue [" .. table.concat(q, ",") .. "]")
end
return innerB(g)
end
love.keypressed(B_KEY, B_KEY, false)
U.log(("D probe: right after keypressed -- queue=[%s] state.b=%s")
:format(table.concat(game.input.pressQueue, ","),
tostring(game.input.state.b)))
U.wait(2)
love.keyreleased(B_KEY, B_KEY)
U.wait(2)
CatchThrow.buttons = innerB
U.log("D probe: buttons saw a non-empty queue " .. seen .. " time(s)")
local ran = false
for _ = 1, 60 do
U.wait(2)
@@ -274,8 +360,7 @@ return function(game)
pcall(CatchThrow.onBattleEnded)
d.result, d.phase, d.afterQueue = "run", "messages", "finish"
end
for _ = 1, 80 do U.tap(game, "a"); U.wait(4) end
U.wait(60)
closeOut(d)
game.speedOverride = nil
end
U.log("done -- " .. DIR)