Declare distributed outputs and clarify scenario authoring

.github/workflows/Benchmarks.yml

@@ -59,7 +59,7 @@ jobs:
         uses: actions/checkout@v7
         with:
           repository: VEZY/PlantBiophysics.jl
-          ref: multi-plant
+          ref: master
           path: downstream/PlantBiophysics
 
       - name: Set up Julia
@@ -106,6 +106,6 @@ jobs:
     if: ${{ github.event_name == 'workflow_dispatch' && inputs.full_performance }}
     uses: ./.github/workflows/FullPerformance.yml
     with:
-      xpalm_ref: codex/xpalm-release-regression
-      plantbiophysics_ref: multi-plant
-      plantgeom_ref: plantsimengine-multi-plant
+      xpalm_ref: main
+      plantbiophysics_ref: master
+      plantgeom_ref: main

.github/workflows/FullPerformance.yml

@@ -21,7 +21,7 @@ on:
       plantgeom_ref:
         description: PlantGeom branch, tag, or commit
         required: true
-        default: plantsimengine-multi-plant
+        default: main
         type: string
   workflow_call:
     inputs:
@@ -38,7 +38,7 @@ on:
       plantgeom_ref:
         description: PlantGeom branch, tag, or commit
         required: false
-        default: plantsimengine-multi-plant
+        default: main
         type: string
 
 permissions:
@@ -84,7 +84,7 @@ jobs:
         uses: actions/checkout@v7
         with:
           repository: VEZY/PlantGeom.jl
-          ref: ${{ inputs.plantgeom_ref || 'plantsimengine-multi-plant' }}
+          ref: ${{ inputs.plantgeom_ref || 'main' }}
           path: downstream/PlantGeom
 
       - name: Set up Julia

AGENTS.md

@@ -10,6 +10,8 @@ API.
 - `@process` defines the abstract process type.
 - `process(model)` identifies the process.
 - `inputs_(model)` and `outputs_(model)` declare status variables.
+- In `outputs_`, `Distributed(Default(value))` or `Distributed(Required(T))`
+  declares an output on selected destination objects. Ordinary values stay local.
 - `environment_inputs_(model)` and `environment_outputs_(model)` declare environment
   variables.
 - `dep(model)` optionally returns model-author defaults using `Input(...)` and
@@ -27,8 +29,9 @@ Read the current model's parameters directly from `model`. `context` is a
 
 - `CompositeModel` owns a `ObjectRegistry`, model applications, instances, and an
   environment.
-- `Object` is one runtime entity with stable `ObjectId`, labels, parent,
-  geometry, and `Status`.
+- `Object` is one runtime entity with a unique, stable `ObjectId`, optional
+  display `name`, group labels, parent, geometry, and `Status`. Display names
+  need not be unique and never select objects; the UI shows the ID by default.
 - Plant architecture is not prescribed. Users choose scales and topology.
 - `CompositeModelTemplate` and `ObjectInstance` reuse the same model definitions across
   several plants or objects.
@@ -47,6 +50,10 @@ ModelSpec(model; name=:application, on=selector, inputs=(...), calls=(...), ever
 
 - `on` selects where the model runs.
 - `inputs` declares value dependencies.
+- `outputs_to` is a tuple of anonymous `OutputTo(selector; vars=...)` entries.
+  One entry can omit `vars` to bind all distributed outputs; several entries
+  must explicitly partition their names. Kernels request columns with
+  `output_targets(context, (:variable,))`.
 - `calls` declares manually executable hard dependencies.
 - `Updates(:x; after=:producer)` orders intentional duplicate writers.
 - `output_routing=(x=:stream_only,)` excludes an output from canonical
@@ -67,11 +74,15 @@ Scope and topology:
 - `Subtree()`: the current object and its descendants
 - `SelfPlant()`: the current plant instance/root
 - `Ancestor(...)`
-- `Scope(name)`
+- `Scope(:instance_name)`: an instance's root and descendants
+- `Scope(ObjectId(root_id))`: a root chosen directly by object ID and its descendants
 - `Relation(...)`
 
-Use keyword criteria for object labels: `kind=:plant`, `species=:oil_palm`,
-`scale=:Leaf`, and `name=:leaf_1`.
+Use `id=:leaf_1` to select a particular object. Preserve the ID's value type,
+including numeric MTG IDs. Use keyword criteria for group labels:
+`kind=:plant`, `species=:oil_palm`, and `scale=:Leaf`.
+`ModelSpec.name` and `ObjectInstance.name` remain identifiers for applications
+and instances; do not treat them as object display names.
 
 `Self()` never means the model, species, or plant unless the current object is
 itself that plant.

CHANGELOG.md

@@ -1,5 +1,17 @@
 # Changelog
 
+## Unreleased
+
+- Declare outputs written to selected objects inside `outputs_` with
+  `Distributed(Default(value))` or `Distributed(Required(T))`. These variables
+  are now visible when inspecting a model before building a scenario.
+- `outputs_to` now takes an anonymous tuple of `OutputTo` entries. One entry
+  can omit `vars` to bind all distributed outputs; multiple entries must
+  explicitly partition variable names. Defaults and types belong to the model.
+- Request destination columns with `output_targets(context, (:x, :y))`.
+  Missing, duplicate, and incompatible destination declarations fail before
+  initialization. Named output groups and schemas in `OutputTo.vars` were removed.
+
 ## v0.15.0
 
 ### Summary

Project.toml

@@ -24,7 +24,7 @@ HTTP = "cd3eb016-35fb-5094-929b-558a96fad6f3"
 PlantSimEngineGraphEditorExt = "HTTP"
 
 [compat]
-CSV = "0.10"
+CSV = "0.10, 1.0"
 DataFrames = "1"
 Dates = "1.10"
 HTTP = "1, 2.0"

README.md

@@ -40,6 +40,12 @@ ModelSpec(
 This is the package API for multiscale, multi-plant, soil, microclimate, and
 model-scale simulations.
 
+An object needs a unique ID, for example `Object(:leaf_1)`. Select it with
+`One(id=:leaf_1)`, or add `scale=:Leaf` to select it as part of
+`Many(scale=:Leaf)`. An optional `name="Sunlit leaf"` changes its display
+label; it does not change the object's identity or connections. Model
+application names identify the configured uses of models.
+
 ## Installation
 
 The examples on this development branch use `CompositeModel`. Registered

benchmark/Project.toml

@@ -3,17 +3,13 @@ BenchmarkTools = "6e4b80f9-dd63-53aa-95a3-0cdb28fa8baf"
 CSV = "336ed68f-0bac-5ca0-87d4-7b16caf5d00b"
 DataFrames = "a93c6f00-e57d-5684-b7b6-d8193f3e46c0"
 MultiScaleTreeGraph = "dd4a991b-8a45-4075-bede-262ee62d5583"
-PlantBiophysics = "7ae8fcfa-76ad-4ec6-9ea7-5f8f5e2d6ec9"
 PlantMeteo = "4630fe09-e0fb-4da5-a846-781cb73437b6"
 PlantSimEngine = "9a576370-710b-4269-adf9-4f603a9c6423"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 SHA = "ea8e919c-243c-51af-8825-aaa63cd721ce"
 Sockets = "6462fe0b-24de-5631-8697-dd941f90decc"
 Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
-XPalm = "6b523e1e-d512-416c-8e51-a8fbef0064e7"
 
 [sources]
 PlantSimEngine = {path = ".."}
-XPalm = {rev = "main", url = "https://github.com/PalmStudio/XPalm.jl"}
-PlantBiophysics = {rev = "master", url = "https://github.com/VEZY/PlantBiophysics.jl"}

benchmark/README.md

@@ -0,0 +1,20 @@
+# Benchmarks
+
+`Project.toml` contains the dependencies for the PlantSimEngine benchmarks.
+The default suite runs only these benchmarks, both locally and in CI.
+AirspeedVelocity resolves this project before loading `benchmarks.jl`, so
+optional downstream packages must stay out of its dependencies and sources.
+
+The dedicated PlantBiophysics and full downstream workflows add their checked-out
+packages with `Pkg.develop` after running the corresponding helper in
+`prepare_full_performance_project.jl`. The full workflow develops XPalm,
+PlantBiophysics, PlantGeom, and the PlantSimEngine checkout together, so XPalm
+uses the compatible PlantGeom source instead of an older registry release.
+
+For a local downstream run, follow the environment setup in
+`.github/workflows/FullPerformance.yml`, then set
+`PSE_BENCHMARK_INCLUDE_DOWNSTREAM=true` when loading `benchmarks.jl`.
+The flag enables the downstream suite; it does not install its dependencies.
+`test/runtests.jl` uses the same default. Passing a test-name pattern explicitly
+selects those tests, as the dedicated downstream CI jobs do.
+The pinned release comparisons in `release_baselines/` use separate projects.

benchmark/benchmarks.jl

@@ -20,7 +20,7 @@ const SUITE = BenchmarkGroup()
 const INCLUDE_DOWNSTREAM_BENCHMARKS = get(
     ENV,
     "PSE_BENCHMARK_INCLUDE_DOWNSTREAM",
-    get(ENV, "GITHUB_ACTIONS", "false") == "true" ? "false" : "true",
+    "false",
 ) == "true"
 _supports_composite_object_benchmarks(engine) =
     isdefined(engine, :CompositeModel) &&

benchmark/initial_status_preparation.jl

@@ -100,7 +100,7 @@ function readout_binding(name, j, width)
         return name => One(within=Self(), var=Symbol("seed_", mod1(j, width)))
     elseif kind == 2
         return PreviousTimeStep(name) => One(
-            scale=:Source, name=:source_1, within=Subtree(),
+            scale=:Source, id=:source_1, within=Subtree(),
             application=:signal_source, var=:signal,
         )
     end

benchmark/prepare_full_performance_project.jl

@@ -3,6 +3,10 @@ using TOML
 function prepare_full_performance_project!(project_path)
     project = TOML.parsefile(project_path)
     pop!(project, "sources", nothing)
+    # Downstream packages are opt-in: AirspeedVelocity resolves the core project
+    # before benchmarks.jl can inspect PSE_BENCHMARK_INCLUDE_DOWNSTREAM.
+    project["deps"]["XPalm"] = "6b523e1e-d512-416c-8e51-a8fbef0064e7"
+    project["deps"]["PlantBiophysics"] = "7ae8fcfa-76ad-4ec6-9ea7-5f8f5e2d6ec9"
     open(project_path, "w") do io
         TOML.print(io, project)
     end
@@ -13,6 +17,7 @@ function prepare_plantbiophysics_performance_project!(project_path)
     project = TOML.parsefile(project_path)
     pop!(project, "sources", nothing)
     pop!(project["deps"], "XPalm", nothing)
+    project["deps"]["PlantBiophysics"] = "7ae8fcfa-76ad-4ec6-9ea7-5f8f5e2d6ec9"
     open(project_path, "w") do io
         TOML.print(io, project)
     end

benchmark/test-distributed-output-benchmark.jl

@@ -9,8 +9,12 @@ struct DistributedOutputBenchmarkBoundInputModel <:
        AbstractDistributed_Output_Benchmark_Bound_InputModel end
 struct DistributedOutputBenchmarkStatusInputModel <:
        AbstractDistributed_Output_Benchmark_Status_InputModel end
-struct DistributedOutputBenchmarkSceneWriterModel <:
-       AbstractDistributed_Output_Benchmark_Scene_WriterModel end
+struct DistributedOutputBenchmarkSceneWriterModel{D} <:
+       AbstractDistributed_Output_Benchmark_Scene_WriterModel
+    declarations::D
+end
+DistributedOutputBenchmarkSceneWriterModel() =
+    DistributedOutputBenchmarkSceneWriterModel(NamedTuple())
 struct DistributedOutputBenchmarkAssignmentModel{T,I,C,M} <:
        AbstractDistributed_Output_Benchmark_AssignmentModel
     table::T
@@ -49,9 +53,12 @@ PlantSimEngine.outputs_(::DistributedOutputBenchmarkStatusInputModel) = (
     total=0.0,
 )
 PlantSimEngine.inputs_(::DistributedOutputBenchmarkSceneWriterModel) = NamedTuple()
-PlantSimEngine.outputs_(::DistributedOutputBenchmarkSceneWriterModel) = NamedTuple()
+PlantSimEngine.outputs_(model::DistributedOutputBenchmarkSceneWriterModel) = model.declarations
 PlantSimEngine.inputs_(::DistributedOutputBenchmarkAssignmentModel) = NamedTuple()
-PlantSimEngine.outputs_(::DistributedOutputBenchmarkAssignmentModel) = NamedTuple()
+PlantSimEngine.outputs_(model::DistributedOutputBenchmarkAssignmentModel) =
+    NamedTuple{keys(model.columns)}(map(values(model.columns)) do column
+        Distributed(Default(zero(eltype(column))))
+    end)
 function PlantSimEngine.run!(
     ::DistributedOutputBenchmarkSceneWriterModel,
     status,
@@ -69,7 +76,7 @@ function PlantSimEngine.run!(
     constants,
     context,
 )
-    targets = PlantSimEngine.output_targets(context, :leaves)
+    targets = PlantSimEngine.output_targets(context, keys(model.columns))
     benchmark_assign_outputs_api!(model.mode, targets, model)
     return nothing
 end
@@ -448,9 +455,6 @@ function setup_distributed_output_public_assignment_benchmark(
            data.column_values.rank :
            reverse(data.column_values.rank)
     table = ncolumns == 1 ? base_table : (; base_table..., rank)
-    output_variables = ncolumns == 1 ?
-                       (incident_par=Default(0.0),) :
-                       (incident_par=Default(0.0), rank=Default(0))
     objects = Object[Object(:scene; scale=:Scene)]
     sizehint!(objects, nobjects + 1)
     for (index, object_id) in enumerate(data.object_ids)
@@ -481,9 +485,8 @@ function setup_distributed_output_public_assignment_benchmark(
                 name=:scene_assignment,
                 on=One(scale=:Scene),
                 outputs_to=(
-                    leaves=OutputTo(
-                        Many(scale=:Leaf, within=SceneScope());
-                        vars=output_variables,
+                    OutputTo(
+                        Many(scale=:Leaf, within=SceneScope()),
                     ),
                 ),
             ),
@@ -503,9 +506,6 @@ function setup_distributed_output_wide_assignment_benchmark(
     columns = NamedTuple{names}(
         ntuple(index -> fill(Float64(index), nobjects), ncolumns),
     )
-    output_variables = NamedTuple{names}(
-        ntuple(_ -> Default(0.0), ncolumns),
-    )
     objects = Object[Object(:scene; scale=:Scene)]
     sizehint!(objects, nobjects + 1)
     for object_id in data.object_ids
@@ -532,9 +532,8 @@ function setup_distributed_output_wide_assignment_benchmark(
                 name=:scene_wide_assignment,
                 on=One(scale=:Scene),
                 outputs_to=(
-                    leaves=OutputTo(
-                        Many(scale=:Leaf, within=SceneScope());
-                        vars=output_variables,
+                    OutputTo(
+                        Many(scale=:Leaf, within=SceneScope()),
                     ),
                 ),
             ),
@@ -625,13 +624,14 @@ function setup_distributed_output_compilation_benchmark(
     end
     application = if distributed
         ModelSpec(
-            DistributedOutputBenchmarkSceneWriterModel();
+            DistributedOutputBenchmarkSceneWriterModel((
+                incident_par=Distributed(Default(0.0)),
+            ));
             name=:scene_writer,
             on=One(scale=:Scene),
             outputs_to=(
-                leaves=OutputTo(
-                    Many(scale=:Leaf, within=SceneScope());
-                    vars=(incident_par=Default(0.0),),
+                OutputTo(
+                    Many(scale=:Leaf, within=SceneScope()),
                 ),
             ),
         )