R.maps.objects was f.opt(f.list(f.any)): a static wild encounter's
species (OverworldController.lua's d.pokemon, handed straight to
BattleState.newWild) went completely unchecked at load time, unlike an
encounter slot's species. A typo'd or removed id sat in a loaded mod
and only surfaced as a crash the moment a player reached that object.
Objects share one array across every kind -- NPCs, signs, warps and
static encounters all coexist with no field the loader could use to
tell them apart ahead of time -- so a strict f.rec covering the whole
shape would reject every kind this schema does not enumerate. Added
f.partial, an open counterpart to f.rec: it type-checks (and, through
collectRefs, cross-reference-checks) only the fields it is given and
leaves everything else on the value alone, the same extensibility
f.rec already grants at a record's top level but nowhere further in.
R.maps.objects now types just `pokemon` through it, so a bad species
id is a load-time "unresolved reference" error instead of a runtime
crash, while an NPC object's sprite/movement/range/... fields -- never
named in this schema -- still pass through untouched.
ItemEffects.use walked speciesDef.tmhm with a bare ipairs() to check
whether the species could learn the machine's move. A record with no
tmhm field at all -- a mod species that never set one, or any record
missing it for whatever reason -- hit ipairs(nil) and took the whole
game down on the first TM/HM use, rather than reaching the ordinary
"can't learn that move" refusal a species whose list simply omits the
move already gets.
An absent list now reads the same as an empty one: nothing to learn,
same refusal, same sound, same text.
useOn was a plain Lua local: every result ItemEffects.use returned fell
through to one unconditional showMessages with no seam a mod could
reach, unlike menu.lua/boxmark.lua/formview.lua's screens, which wrap
their own default behavior as a table field or a Runtime hook. A mod
could not suppress a message, delay it behind a screen of its own, or
substitute a different outcome for one item id -- exactly the gap noted
against Ultra Burst's item-driven fusion, which had nowhere left to
attach a bespoke animation once TextBox.new turned out to be the only
other reachable seam.
This wraps the whole dispatch in a Runtime.call("item.use", ...) hook,
the same mechanism "battle.overlay", "ui.party.submenu" and the rest of
src/ui already use, rather than exporting BagMenu.useOn as a table
field. A hook is the smaller commitment: it is additive (a fresh
Runtime.call site needs no schema or manifest change and costs nothing
unsubscribed -- see tests/engine/gate_hooks.lua's null-object case) and
a mod can still run the vanilla flow unchanged by calling the handed-in
vanilla function, whereas a table field would fix useOn's exact
signature as public API the moment it shipped. If the maintainer would
rather match the sibling screens' convention directly, exporting
BagMenu.useOn is the alternative and does not conflict with this hook
existing alongside it.
vanillaUseOn keeps the original function body; useOn is now the thin
wrapper mods observe through, and every internal caller in this file
still goes through useOn so the hook fires on every path into it.
stepHPDrain counts drainHold down to 0 as the last step of every phase
(pixel slide, HP-number step, closing frames) but never let go of the
field afterward, so it sat at 0 -- not nil -- for the rest of the
battle. BattleSafety.inspect uses drainHold ~= nil as its
settled-presentation gate for checkpoint capture, so the very first HP
change in a battle permanently refused every checkpoint after it with
battle_phase_busy, even once the bar had long since caught up.
Only nil the field when the whole drain is actually over (bar pixel,
HP number and the closing-frame hold all settled), not on every
mid-sequence 0 -- a fresh HP change still needs drainHold to read as
busy so BattleSafety keeps refusing captures until that one settles
too.
tmpnam() on the Windows CRT returns a bare, CWD-relative name (e.g. \sb4c.2), and io.open on it fails with Permission denied when the game's working directory is not writable -- a Program Files (or otherwise protected) install. postLog then dies before curl runs: the mod reports a send failure and no bytes leave the machine (confirmed on a Windows install: "could not create request body: \sb4c.2: Permission denied").
Stage the request body under the OS temp contract instead: TEMP/TMP on Windows (always set, always per-user writable), TMPDIR with a /tmp fallback on POSIX. The transport stays on plain io/os -- no love.filesystem dependency.
Tests updated to mock os.getenv and assert the staged path sits under the temp dir; 10/10 checks pass.
The save editor now routes Gold vs RBY through a generation adapter so boxes, items, events, and maps write the right fields. The launcher layout is shuffled a bit, and patch notes can come from the updater, a packed file, or the iOS sidecar.
Regression coverage for the four Yellow Pallet Town fixes on this
branch:
- tests/parity_J.lua: the old-man-style demo bag shows x1 in Yellow
(SimulatedInputBattleItemList), not just the pre-existing x50
(pokered's OldManItemList) case.
- tests/engine/push_battle_transition.lua: Commands.pushBattle calls
ctx.overworld:pushBattle when available and falls back with a
logged warning otherwise. Logger.warn is spied (pcall-safe, always
restored) rather than read off the shared Logger.history ring
buffer, so the fallback checks don't leave noise behind for
whatever else runs in the same process.
- tests/parity_yellow_pallet_pikachu.lua: drives the real onStep
closure through the real StateStack/OverworldState, with no mocked
battle -- Red never plays Music_MuseumGuy for this escort (Blue
shares the same code path: onStep only branches on
GameVersion.isYellow(), never isBlue(), so a separate Blue run
would exercise nothing new), and Yellow's hold before the Pikachu
battle is armed for exactly 2 frames before handing off to
BattleTransition rather than a bare stack push. Stops there rather
than also driving the demo battle to completion just to assert
Music_MuseumGuy fires in Yellow (that fix's own coverage): the
extra coupling to unrelated battle menu/bag/throw frame budgets
wasn't worth it for one more assertion. That side stays manually
verified. scenario() restores Game/GameVersion and re-inits
StateStack on the way out.
All three pass standalone (luajit tests/<path>). The Yellow E2E test
hits the same pre-existing "Music.playMap is nil" gap three other
map-warping parity tests already hit inside the aggregated
tests/run_tests.lua run (missing data/generated/audio.lua in this dev
environment) -- confirmed by diffing the identical error text against
parity_warp_after_warp_step.lua, which also passes clean standalone.
sanjinpepic
AverageConsumer
bryanthaboi
Shane McGovern
anxiousintrovert
1jamie
MaxTomahawk
bryanthaboi
Codex Agent
thibautbus
MaxTomahawk