ConstructionSystem takes over the construction queue: it runs the front site's
timer and, when it elapses, builds the Building itself — ports, buffers, belt
registration, and starting the next queued site. Simulation::tick calls it
directly, in the same position tickConstruction held.
No handoff. The earlier sketch had it return the completed site for BuildingSystem
to materialise, which put an intermediate value in Simulation and made the two
calls correct only when adjacent and ordered. Once the queries and the buffer
helpers became free functions there was nothing left on BuildingSystem that
materialisation needed, so the system does the whole job and the invariant
disappears rather than being documented.
Holds only the config; the world arrives per tick, like the systems in
lib/ecs/system.
Verified with a golden-checksum capture before and after — all four sample ticks
identical, which is the check that matters here since this moves a call in the
tick order.
Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YHcUerKAZKWNvSKJxYKbnG
Prerequisite for ConstructionSystem completing a building itself rather than
handing a finished site back: materialisation needs the buffer initialisers and
the BeltSystem registration, and both were BuildingSystem members.
initBuffers turned out to need nothing at all — it works purely on the Building
and RecipeDef it is given. The other three need only the config.
reregisterBeltTile takes BeltSystem and the config; it stays shared rather than
moving, because cancelDeconstruction and rotateInPlace use it too and are staying
on BuildingSystem.
Verified with a golden-checksum capture before and after — all four sample ticks
identical.
Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YHcUerKAZKWNvSKJxYKbnG
The member reference is gone. All 23 methods that read or write the factory now
take FactoryState& (const for the two item walks and the checksum fold), so a
BuildingSystem is no longer bound to one state and its signatures say which data
each call touches. It holds only config, belts, rng and the callbacks — the same
shape as AiSystem and CombatSystem.
This completes what phase 2 set out to do; the ownership move landed earlier, but
the systems kept reaching the data through a member until the queries were off
them.
Seeding the asteroid bound moved with the state, and that broke four tests: the
fixtures build their own FactoryState, which defaulted the bound to 0 and refused
every placement on the asteroid. Rather than fix the four call sites, makeFactoryState()
now creates a run's state from the config, and Simulation, ArenaSimulation and the
test fixtures all use it — there is one place that knows what a fresh factory
looks like.
Verified with a golden-checksum capture before and after — all four sample ticks
identical — and by re-running the declaration/definition check over the header.
Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YHcUerKAZKWNvSKJxYKbnG
isPlacementValid, findRotateInPlaceTarget and the bodyCellsWithinWorldBounds
helper become PlacementRules.h — where a building may go and what already sits
on those tiles, answered from the factory state and the config. getInputPorts
and getSiteSplitterInfo join FactoryQueries.h, whose header comment now says
plainly that the last two also take the config because answering them means
reading a building definition.
computeInputPorts goes to PortGeometry.h alongside outputBodyTile/inputBodyTile:
it needs only Port and QPoint, so it belongs in core rather than in sim.
BuildingSystem is left with no query that reads the factory — its remaining const
methods are the emerging/incoming item walks, the checksum fold, and the buffer
initialisers. It changes the factory now; it no longer describes it.
Verified with a golden-checksum capture before and after — all four sample ticks
identical.
Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YHcUerKAZKWNvSKJxYKbnG
gatherCandidateRecipes, recipeInputsAvailable, computeShipyardRequiredMaterials,
hasInputsToStart and getProductionStatus read no factory state — they answer
"what can this building produce, and can it start" from the config and the
Building alone. They move to ProductionRules.h as free functions, and the
ProductionStatus enum goes with them since it is that group's return type.
Two of the five are pure in their arguments; the other three need GameConfig
through gatherCandidateRecipes and computeShipyardRequiredMaterials, so config is
a parameter rather than the group being split across two headers.
Only two callers outside BuildingSystem existed — the status light in
GameWorldView and one test lambda — so this is nearly all internal.
Verified with a golden-checksum capture before and after — all four sample ticks
identical.
Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YHcUerKAZKWNvSKJxYKbnG
The eleven forwarding members added last commit are gone; callers now read the
data directly through FactoryQueries.h. Simulation and ArenaSimulation expose
getFactoryState() so the UI, the balancing view and the tests can reach it.
isQueuedForDeconstruction joined the free functions along the way — it only
reaches findBuilding, so it was state-pure too.
No facade was introduced. The chained form was the reason one looked attractive,
but rewriting sim.getBuildings().findBuilding(id) to findBuilding(sim.getFactoryState(), id)
turned out to be mechanical, and the result says which data is read rather than
which system happens to own it.
BuildingSystem.cpp is down to 1735 lines and no longer answers questions about
the factory — it only changes it. What remains on it are the mutators, the tick
phases, and the queries that also need GameConfig.
Verified with a golden-checksum capture before and after — all four sample ticks
identical.
Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YHcUerKAZKWNvSKJxYKbnG
It was the last piece of mutable world data BuildingSystem still owned, and the
placement queries need it: isPlacementValid reaches it through
bodyCellsWithinWorldBounds, so those queries cannot become free functions over
FactoryState while the bound lives on the system.
Left out of the checksum deliberately. It is derived from config and Simulation's
expansion count, which is folded already, so adding it would change every
checksum without adding information.
Still seeded from config by BuildingSystem's constructor, which keeps the
initialization at exactly the point it happened before; reset() clears the state
before initializeSubsystems() rebuilds the system, so the ordering holds.
Verified with a golden-checksum capture before and after — all four sample ticks
identical.
Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YHcUerKAZKWNvSKJxYKbnG
Ten queries that read nothing but FactoryState become free functions in
FactoryQueries.h; the BuildingSystem methods stay as one-line forwards, so no
existing caller moves yet.
That lets the AI path drop its dependency on the system entirely. AiSystem,
SalvagerSystem, DeliverScrapEvaluator and DeliverScrapExecutor took a
BuildingSystem& purely to call findBuilding, findNearestBuilding and
deliverScrapToSalvageBay — all three are state-pure — so they now take
FactoryState& and say what they actually read. Four forward declarations of
BuildingSystem go with them.
No facade: the queries are plain free functions over the data. A facade was
considered to spare the ~180 UI call sites, but the AI needed only the data and
would have been given GameConfig it has no use for.
isProductionBuildingType moves to BuildingType.h beside isAutoRecipeBuildingType
and isBeltSubsystemType rather than being copied into the new file.
Verified with a golden-checksum capture before and after — all four sample ticks
identical.
Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YHcUerKAZKWNvSKJxYKbnG
Simulation (and ArenaSimulation in the balancing tool) now owns the factory's
world data; BuildingSystem holds a reference to it. This is what lets the systems
that operate on the data be handed the same state — phase 3's construction and
deconstruction systems, and later the ecs/system/ classes that today take a
BuildingSystem& only to query it.
reset() clears the state alongside m_admin and m_beltSystem, matching how the
subsystems were already rebuilt from scratch.
Falls short of the tick-argument form I sketched: BuildingSystem still reaches
the data through a member reference rather than a parameter. Making it truly
stateless means the const query surface has to find the data some other way, and
that surface is large — findBuilding alone has 64 call sites, with findSite,
getAllBuildings, getAllSites, isTileOccupied and the rest behind it. Doing that
needs a queries facade behind Simulation::getBuildings() so the callers do not
all move, which is its own decision rather than a side effect of this one.
The constructor gains a parameter, so the four owners and the three test fixtures
that build a BuildingSystem directly are updated; the 33 files that only use one
are untouched.
Verified with a golden-checksum capture before and after — all four sample ticks
identical.
Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YHcUerKAZKWNvSKJxYKbnG
BuildingSystem is the one system in the codebase that owns the world data it
operates on. The ecs/system/ classes already do the opposite — AiSystem and
SalvagerSystem hold config and their own scratch, and take EntityAdmin and the
other systems as tick arguments — so this is bringing the outlier in line, not
inventing a pattern.
Phase 1 of that: the buildings vector, both work queues and the tile grid move
into a FactoryState struct, still owned by BuildingSystem. Every method reaches
through m_state. The public API is untouched, so none of the 33 files that
reference BuildingSystem needed a change.
DeconstructionEntry moves out of BuildingSystem's private section into
FactoryState.h, since the queue that holds it lives there now.
m_asteroidWidth_tiles stays on the system: it is not checksummed and is a cached
placement bound derived from config and the expansion count, not factory data.
The intent is for Simulation to own FactoryState and pass it into the tick
methods, leaving the systems stateless over it. FactoryState.h notes explicitly
that this is a data/behaviour split and not a step toward putting buildings in
the entity model, which architecture.md rules out.
Verified with a golden-checksum capture before and after — all four sample
ticks identical.
Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YHcUerKAZKWNvSKJxYKbnG
Eleven methods maintained m_tileOccupancy by hand — place, deconstruct,
removeBuilding, placeImmediate, tickDeconstruction, findRotateInPlaceTarget and
tryDirectCoupleDeposit all indexed a raw std::map<std::pair<int,int>, BuildingId>
directly, so the invariant "occupancy stays in sync with placement" was
re-implemented at every call site. They now ask and tell a small owned index
instead: occupy / release / isOccupied / findOwner.
BuildingGrid is a member of BuildingSystem, not a peer system: it has no
per-tick behaviour and nothing outside BuildingSystem touches it.
The internal keying stays std::pair<int,int> rather than moving to QPoint. The
checksum folds the entries in map iteration order, so the comparator is part of
the determinism contract; changing it is a separate decision, not a side effect
of this move. Verified with a golden-checksum capture before and after — all
four sample ticks identical.
Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YHcUerKAZKWNvSKJxYKbnG
Four of the five file-local helpers in BuildingSystem.cpp had duplicates
elsewhere: isAutoRecipeBuildingType and isBeltSubsystemType were re-spelled as
isAutoRecipeBuilding and isBeltLike in SelectedBuildingPanel.cpp, and
outputBodyTile was copied verbatim as portBodyTile in GameWorldView.cpp. Same
predicates, different names, so a change to one would silently not reach the
others.
The two BuildingType predicates move to BuildingType.h, which already hosts the
free functions over that enum and is already included by both lib and ui. The
port geometry moves to a new PortGeometry.h; inputBodyTile has no duplicate but
is outputBodyTile's counterpart and belongs beside it — the sim moves items
across the port edge and the renderer draws the virtual belt there, so the two
must agree on which tile a port owns.
inputLaneEntryFree stays file-local: single use, and tied to BeltItemSlot rather
than to building types or port geometry.
Verified the six moved bodies are character-identical to their originals.
Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YHcUerKAZKWNvSKJxYKbnG
BuildingSystem::findBuildingDef was a byte-equivalent re-implementation of
BuildingsConfig::findBuildingDef. Same cleanup as the GameWorldView copy;
this one sits in the sim layer, so it was missed by both earlier passes.
Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GH8ZMRY3vhxxXcaUxBqxkk
Ships/Modules/RecipesConfig now carry lookup helpers mirroring
BuildingsConfig::findBuildingDef. The hand-rolled linear scans in
BuildingSystem, ShipSystem, ShipStatsCalculator, ThreatCostCalculator,
ShipLayoutDialog, SelectedBuildingPanel and SchematicChoiceDialog now
call them instead.
Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GH8ZMRY3vhxxXcaUxBqxkk
rotateInPlace re-implemented the belt-tile re-registration switch inline
instead of calling reregisterBeltTile, and its splitter branch omitted the
setSplitterFilters call the canonical version has. Since an operational
splitter keeps its filters only in BeltSystem, removeTile discarded them and
rotating a configured splitter silently reset it to "accept all".
Replace the duplicated switch with a call to reregisterBeltTile, capturing
the filters beforehand via getSplitterInfo — the same idiom deconstruct
already uses. This removes the second copy of the switch that allowed the
two to drift apart in the first place.
Add a regression test; the existing [rotate-in-place] cases covered belt
tiles only, which is why this went unnoticed.
Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GH8ZMRY3vhxxXcaUxBqxkk
Add deterministic record/playback for a run.
Recording captures `(seed, config hash, ordered tick-tagged commands)` and re-simulates on playback — no state snapshots. `DotaFactory.exe --replay <file>` re-plays a recorded run view-only with manual speed/pause.
Reviewed-on: #4
Co-authored-by: Malte Langkabel <malte.langkabel@gmail.com>
Co-committed-by: Malte Langkabel <malte.langkabel@gmail.com>
