Attributes as table

AGENT.md

@@ -21,6 +21,7 @@
 - `src/compute_MTG/indexing.jl`
 - `src/compute_MTG/check_filters.jl`
 - `src/types/Node.jl`
+- `src/types/Attributes.jl`
 - `src/compute_MTG/node_funs.jl`
 
 ## Practical Optimization Rules
@@ -30,4 +31,14 @@
   - `descendants!(buffer, node, key; ...)`
 - Keep filter checks branch-light when no filters are provided.
 - Keep key access on typed attribute containers (`NamedTuple`, `MutableNamedTuple`, typed dicts) in specialized methods when possible.
+- Prefer explicit attribute APIs:
+  - `attribute(node, key)`
+  - `attribute!(node, key, value)`
+  - `attributes(node; format=:namedtuple|:dict)`
+  - `add_column!` / `drop_column!` / `rename_column!`
 - Preserve API behavior and add tests for every optimization that changes internals.
+
+## Benchmark Tiers
+- `small` (~10k nodes): full matrix including API-surface benchmarks
+- `medium` (~100k nodes): hot-path matrix
+- `large` (~300k nodes): critical hot paths only

Project.toml

@@ -14,6 +14,7 @@ MutableNamedTuples = "af6c499f-54b4-48cc-bbd2-094bba7533c7"
 OrderedCollections = "bac558e1-5e72-5ebc-8fee-abe8a469f55d"
 Printf = "de0858da-6303-5e67-8744-51eddeeeb8d7"
 SHA = "ea8e919c-243c-51af-8825-aaa63cd721ce"
+Tables = "bd369af6-aec1-5ad0-b16a-f7cc5008161c"
 XLSX = "fdbf4ff8-1666-58a4-91e7-1b58723a45e0"
 
 [compat]
@@ -25,6 +26,7 @@ MetaGraphsNext = "0.5, 0.6, 0.7"
 MutableNamedTuples = "0.1"
 OrderedCollections = "1.4"
 SHA = "0.7, 1"
+Tables = "1"
 XLSX = "0.7, 0.8, 0.9, 0.10"
 julia = "1.6"
 

benchmark/Project.toml

@@ -1,3 +1,7 @@
 [deps]
 BenchmarkTools = "6e4b80f9-dd63-53aa-95a3-0cdb28fa8baf"
 MultiScaleTreeGraph = "dd4a991b-8a45-4075-bede-262ee62d5583"
+Tables = "bd369af6-aec1-5ad0-b16a-f7cc5008161c"
+
+[sources]
+MultiScaleTreeGraph = {path = ".."}

benchmark/README.md

@@ -10,6 +10,21 @@ julia --project=benchmark benchmark/benchmarks.jl
 
 Workloads currently covered:
 
-- full-tree traversal
-- traversal + data extraction from descendants
-- repeated many-small queries (`children`, `parent`, `ancestors`)
+- tiered datasets: `small` (~10k nodes), `medium` (~100k), `large` (~300k)
+- full-tree traversal and traversal updates
+  - one attribute update on all nodes
+  - multi-attribute update on one symbol (`:Leaf`)
+  - multi-attribute update on mixed symbols (`[:Leaf, :Internode]`)
+- descendants queries
+  - one/many attributes, one symbol
+  - one/many attributes, mixed symbols
+  - `ignore_nothing=true/false`
+  - in-place and allocating variants
+- repeated many-small queries
+  - `children`, `parent`, `ancestors`, `ancestors!`
+  - repeated descendants retrieval with and without in-place buffers
+- API surface (small tier)
+  - insertion/deletion/pruning
+  - `transform!`, `select!`
+  - `symbol_table` / `mtg_table`
+  - `write_mtg`