`fmtk luals-addon <outdir>` treats its argument as the parent and always
appends `factorio`, so the bundle landed in `$RUNNER_TEMP/factorio/factorio`
while the config pointed at `$RUNNER_TEMP/factorio/library`. The factorio
types were never loaded, which is why ci reported 642 undefined-global
and 1204 type-not-found against a tree that is clean locally.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The library path was injected by the workflow, so an editor opened on
this repo could not resolve `modules/clusterio/*`. It now comes from
the environment in both places.
It cannot be written relative to this repo: emmylua does not normalise
`..` in a library path, and an absolute path containing `..` fails the
same way, so the location has to come from outside.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The nil check burn down added guards which turned out redundant once
the surrounding annotations landed. `_has_handlers` is set to true when
the first handler registers, so it is a boolean rather than the literal
false it was inferred as.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
`assert` returns every argument it is given, so `return assert(x, msg)`
was leaking the message as a second return value.
`get_tile` is documented as taking x and y. The async function class
declared `@operator call` with no parameters, which made every async
call look over supplied.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
These were carried over as disabled from the luals config but report
nothing, so they only ever masked a future regression.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Guards now cover the statements that follow them, and a couple of reads
are restructured so the narrowing survives.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The api documents both as a union with a positional array because both
are accepted as input, which makes every read of `.x` or `.left_top`
optional. Everything read back from the game uses the named form, so
the positional variant is removed and we now always write it that way
too.
The spawn area config offsets and the mine depletion search areas were
the only positional writes left, and apply_offset no longer needs its
fallback.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The player list action setter runs during control setup, before any of
the callbacks that use it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Dropdowns are siblings of the button that reads them, and the loop
bounds come from the interface count.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Dropdowns are siblings of the button that reads them, and the loop
bounds come from the interface count.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
spawn_area accepts both position forms on purpose, so it asserts that
one of the two is present rather than narrowing the type.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Color.struct leaves every channel optional, so the rainbow command
declares its own fully populated colour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Rows and selectors are created alongside the machine selector, so they
exist for anything already in the table.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The interfaces are chests and constant combinators, so the inventory,
logistic sections and first circuit section always exist.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
get_player_roles always returns at least the default or root role, so
get_player_highest_role never returns nil, and callers were right to
treat it that way.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
`debug.getinfo` is optional, but these helpers always ask for a frame
which is on the stack.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
A union of table types makes every lookup optional. Separate index
signatures on one class say the same thing without the nil.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Values read back from the game always use the named members, so they
are narrowed once where they enter a function rather than at each use.
`LuaControl.force` is a ForceID union, so reading a force only method
off it needs the cast the rest of the repo already uses.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
`Roles.new_role` returned the result of `error`, which made it nullable,
and `set_permission_group` returned nil for an unknown group, which
broke the config chain with an index error rather than saying so.
Positions and bounding boxes read back from the game always use the
named members, so they are narrowed once where they enter a function.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Locals declared ahead of an assignment inside a callback were inferred
as nil at every use site.
The role event handlers are given a class so the table is not unified
with the other module handler tables.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
`loadstring` was removed in Lua 5.2, `load` takes a string the same way.
Warp gui child lookups are optional, so they use the same assert the
rest of the repo does.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
`LogisticFilter.signal` became `value` in 2.0, so the clear loop broke
on the first slot and outdated signals were left behind.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
emmylua takes self from the owner table, so `@param self ExpRoles.Role`
only applies with dot syntax. Call sites still use a colon.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Autofill stores the whole per entity map against a player.
The readme container is called as a define, so it needs the overload
the other element classes carry.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
`_display_data` is a cache keyed by force name, not a single record.
The auto complete accumulator starts empty, so its fields are optional.
Game commands have no usage beyond their own help text.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The top, left and relative element tables are keyed by the define, but
the duplicate registration assert looked up `define.name`, so it never
fired.
GuiIter.remove_element indexed `registered_scopes` by player index
instead of the scope elements it had just fetched, so it removed
nothing.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
`config.temp_warning_limit` does not exist, the warnings config defines
`script_warning_limit`, and the locale expects it as `__3__/__4__`, so
the command was printing nil.
map_tags is keyed by the force name joined to the tag number, and
custom_color is the Color union.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
emmylua only parses the inline cast as `--[[@as T]]`, the spaced form
`--[[ @as T ]]` is treated as a plain comment, so all 149 of them were
doing nothing.
LuaGuiElement.style is a union because a style name can be assigned to
it, so reading it back needs the cast.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
`data` on an element is the GuiData store keyed by element, not the
value being stored.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
`get_contents` and `targeted_items_pickup`/`_deliver` return lists of
ItemWithQualityCount in 2.0, they are no longer name keyed dictionaries,
so the logistorage stats were indexing by array position.
PreferenceEnum is bidirectional, matching what ExpUtil.enum returns.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The prune loop shadowed its own parameter, so `silo_data[unit_number] =
nil` wrote into the entry instead of removing it from the table.
`unit_number` is optional on LuaEntity but always set for a silo. Note
that `--[[ @as ]]` does not narrow for an index expression, only
`@cast` does.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The registry and the prototype were both bare tables, so every
`datastore:method()` call resolved to nothing.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The gui element locals are declared at the top and assigned far below,
so emmylua inferred them as nil at every use site.
circuit_oc was reassigned from a control behaviour to a logistic
section, which hid every section method behind the wrong type.
allowed_items entries carry heterogeneous optional properties, now
described by a class.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Async ids come from `get_function_name`, which returns a string, but
were documented as numbers.
`setmetatable` in clusterio's compat module loses `LibCompat` on the
module return, so the require is cast.
`GuiData.__index` accepts a `DataKey` but emmylua needs that as an
index signature on the class.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
emmylua deserialises an unknown diagnostic code to a catch all that
matches no checker, so every `invisible`, `nil-check` and
`global-element` suppression was silently doing nothing. Renamed to the
emmylua codes. `name-style-check` has no equivalent and is dropped, it
was already disabled under luals.
The `get_tile` suppressions referenced an api typedef bug from 2024
which no longer reproduces.
need-check-nil is deferred rather than disabled on merit: a third of the
407 findings come from MapPosition and BoundingBox being aliased as
`struct|[double, double]`, so every `entity.position.x` reads as
possibly nil. The rest need per site knowledge of runtime invariants.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The aabb functions all index the named members, so passing the
shorthand [MapPosition, MapPosition] form would error at runtime. The
annotations now say so.
emmylua requires a type argument on a generic alias, luals did not.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Classes contributed to by more than one file now carry the `partial`
attribute. trains.lua declared its command as `ExpCommand_Artillery`,
which was a copy paste.
`Color.0` and `MapPosition.0` became `.struct` in fmtk 2.1.6.
emmylua rejects `@type` on a function statement or a `do` block, so
those are moved onto the local or replaced by `@param` and `@return`.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
fmtk 2.1.4 onwards generates typedefs targeting emmylua that luals
misreads, and every fmtk changelog since 2.1.5 states luals support is
being dropped. luals 3.18.2 is already the latest release, so there is
nothing to bump.
The `--clusterio-modules` plugin has no emmylua equivalent, as emmylua
has no plugin interface, so it is reimplemented as `workspace.moduleMap`.
Those rules rewrite each file's own module path rather than the require
string, hence the reversed direction.
type.patch.lua is obsolete: `raise_event` now takes `LuaEventType` in
the generated typedefs, and `LuaObject` has become `LuaObject.base`.
exp_scenario/.luacheckrc was already dead, luacheck is never invoked.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
FMTK 2.1.4 onwards reshapes its generated typedefs for EmmyLua, which
LuaLS misreads: 2.1.4 changed dicts to table<K,V> (game.surfaces.nauvis
now reports undefined-field) and 2.1.5 changed LocalisedString to the
variadic tuple [string, LocalisedString...], which LuaLS 3.18.2 cannot
match against a plain string. Together those account for all 279
findings CI has reported since 2026-07-13; with 2.1.3 the check is
clean.
Upstream FMTK has announced it is dropping sumneko/LuaLS support in
favour of EmmyLua, so this pin is a stopgap until the check is ported
to emmylua_check.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
