Macro type hierarchy

Commodity Types

print_tree(Commodity)

Commodity
├─ Alumina
├─ Aluminum
├─ AluminumScrap
├─ Bauxite
├─ Biomass
├─ CO2
│  └─ CO2Captured
├─ Cement
├─ Coal
├─ CrudeSteel
├─ Electricity
├─ Graphite
├─ Heat
├─ Hydrogen
├─ IronOre
├─ LiquidFuels
├─ Ammonia
├─ Methanol
├─ Nitrogen
├─ NaturalGas
├─ Steam
├─ SteelScrap
└─ Uranium

Asset Types

print_tree(AbstractAsset)

AbstractAsset
├─ AluminaPlant
├─ AluminumRefining
├─ AluminumSmelting
├─ BECCSElectricity
├─ BECCSGasoline
├─ BECCSHydrogen
├─ BECCSLiquidFuels
├─ BECCSNaturalGas
├─ Battery
├─ BlastFurnaceBasicOxygenFurnace
├─ BlastFurnaceBasicOxygenFurnaceCCS
├─ CO2Injection
├─ ConstrainedFossilLiquidFuels
├─ DirectReductionElectricArcFurnace
├─ DirectReductionElectricArcFurnaceCCS
├─ DownstreamEmissions
├─ ElectricArcFurnace
├─ ElectricDAC
├─ ElectricHeating
├─ ElectricSteam
├─ Electrolyzer
├─ FuelCell
├─ GasStorage
├─ HydroRes
├─ CementPlant
├─ SyntheticAmmonia
├─ SyntheticMethanol
├─ ThermalAmmonia
├─ ThermalAmmoniaCCS
├─ ThermalMethanol
├─ ThermalMethanolCCS
├─ MustRun
├─ NaturalGasDAC
├─ OneWayTransmissionLink
├─ SyntheticLiquidFuels
├─ SyntheticNaturalGas
├─ ThermalHeating
├─ ThermalHydrogen
├─ ThermalHydrogenCCS
├─ ThermalPower
├─ ThermalPowerCCS
├─ ThermalSteam
├─ TransmissionLink
├─ UpstreamEmissions
└─ VRE

Constraint Types

print_tree(MacroEnergy.AbstractTypeConstraint)

AbstractTypeConstraint
├─ OperationConstraint
│  ├─ BalanceConstraint
│  ├─ CapacityConstraint
│  ├─ LongDurationStorageChangeConstraint
│  ├─ LongDurationStorageImplicitMinMaxConstraint
│  ├─ MaxNonServedDemandConstraint
│  ├─ MaxNonServedDemandPerSegmentConstraint
│  ├─ MaxStorageLevelConstraint
│  ├─ MinDownTimeConstraint
│  ├─ MinFlowConstraint
│  ├─ MinStorageLevelConstraint
│  ├─ MinStorageOutflowConstraint
│  ├─ MinUpTimeConstraint
│  ├─ MustRunConstraint
│  ├─ PolicyConstraint
│  │  ├─ AggregatedDemandConstraint
│  │  ├─ CO2CapConstraint
│  │  └─ CO2StorageConstraint
│  ├─ RampingLimitConstraint
│  ├─ StorageCapacityConstraint
│  ├─ StorageChargeLimitConstraint
│  ├─ StorageDischargeLimitConstraint
│  └─ StorageSymmetricCapacityConstraint
└─ PlanningConstraint
   ├─ AgeBasedRetirementConstraint
   ├─ MaxCapacityConstraint
   ├─ MaxInitStorageLevelConstraint
   ├─ MaxNewCapacityConstraint
   ├─ MinCapacityConstraint
   ├─ MinInitStorageLevelConstraint
   ├─ StorageChargeDischargeRatioConstraint
   ├─ StorageMaxDurationConstraint
   └─ StorageMinDurationConstraint

JuMP container types

Macro's variables and constraints are stored on edges, vertices and storage units as JuMP containers, and the concrete container type is not the same in every model. The types involved are defined in src/model/jump_containers.jl and come in two layers.

The wide types, JuMPVariable and JuMPConstraint, are unions over every container JuMP might hand back (Array, DenseAxisArray, SparseAxisArray, and VariableRef/ConstraintRef). Struct fields are declared with them — flow::JuMPVariable on an Edge, non_served_demand::JuMPVariable on a Node — because a field has to accept whichever container the model build actually produced. array_container decides which one: a contiguous one-based index set such as 1:24 gets a plain Array, anything else gets a DenseAxisArray. In practice the Array fast path is what a monolithic model uses everywhere; the DenseAxisArray branch exists for Benders subproblems, where generate_decomposed_system gives each subproblem its own subperiod window (1:168, 169:336, 337:504, …) and only the first one starts at 1.

The narrow types — VarArrayOrDense, MatrixVarOrDense, AffExprArrayOrDense — are two-member unions covering only the container shapes that can actually occur for a given field, and they are what the accessors re-assert down to:

flow(e::AbstractEdge, t::Int64) = (flow(e)::VarArrayOrDense)[t]
storage_level(g::AbstractStorage, t::Int64) = (storage_level(g)::VarArrayOrDense)[t]
non_served_demand(n::Node, s::Int64, t::Int64) = (non_served_demand(n)::MatrixVarOrDense)[s, t]

Without that assertion the compiler knows only the field's declared type, JuMPVariable, which is too wide to have a fixed layout. Each element read inside a per-timestep loop then has to go through a generic getindex that returns its result as a freshly allocated, tagged heap object rather than as a value in registers. With the assertion, Julia splits the union into its two concrete branches and specializes the read for each, so it produces a VariableRef/AffExpr directly and allocates nothing.

test/test_container_types.jl pins each alias against a real macro-built container and test/test_accessor_allocation.jl asserts the accessors allocate nothing, so a shape that drifts fails a test instead of quietly reverting to Any.