·
assettype · drop-in viewer
mesh.assetType
A mesh is renderable geometry — vertices and triangle indices, and, for a skinned mesh, the per-vertex skinning data that binds it to a skeleton. A component that draws geometry references a mesh, and the renderer binds the mesh's GPU resource for that entity.
byzero-proxy @ DESKTOP-DB3UJOJ·posted 3mo ago
What it does
mesh
A mesh is renderable geometry — vertices and triangle indices, and, for a skinned mesh, the per-vertex skinning data that binds it to a skeleton. A component that draws geometry references a mesh, and the renderer binds the mesh's GPU resource for that entity.
New to how a mesh exists as an asset, as CPU-side vertex data, and as a live GPU resource (and how
renderer.mesh.*moves between them)? Readcore/resource-modelfirst.
Geometry is stored in data.zmsh; the GPU resource is keyed by
the asset's guid.
Create
asset.create("mesh", name, opts) writes the geometry and registers the mesh.
opts is raw geometry — flat arrays; positions and indices are required,
normals and uvs optional:
asset.create("mesh", "tri", {
positions = { 0,0,0, 1,0,0, 0,1,0 }, -- [x,y,z, ...] (required)
indices = { 0, 1, 2 }, -- [i0,i1,i2, ...] (required, 0-based)
normals = { 0,0,1, 0,0,1, 0,0,1 }, -- [x,y,z, ...] (optional)
uvs = { 0,0, 1,0, 0,1 }, -- [u,v, ...] (optional)
})
The full authorable geometry, all flat arrays unless noted:
| Field | Shape | Notes |
|---|---|---|
positions | [x,y,z, ...] | required |
indices | [i0,i1,i2, ...] | required; 0-based, counter-clockwise for the front face (see Conventions) |
normals | [x,y,z, ...] | optional; the surface's outward direction |
uvs | [u,v, ...] | texture UVs, optional; (0,0) is the image's top-left |
colors | [r,g,b,a, ...] | optional |
tangents | [x,y,z,w, ...] | optional (w = handedness) |
uvs1 | [u,v, ...] | the lightmap UV set, optional |
unwrapUvs | boolean | generate uvs1 when it is absent |
skinning | { joints, weights } | per-vertex skin binding (four each per vertex) |
skins | array of skin tables | the skeleton (see renderer.mesh.decode for the shape) |
morphTargets | array of { name?, positions, normals? } | the shapes the mesh blends towards, each a per-vertex displacement |
encode / decode round-trip every one of these, so a skinned mesh can be
built in engine, not only imported. Pass unwrapUvs = true to generate a lightmap
UV set at creation time.
Conventions
Geometry is read against these whether it arrived through asset.create,
renderer.mesh.create, or an importer.
| Convention | How the engine reads the geometry |
|---|---|
| Index base | indices count vertices from 0 — index 0 is the first positions triple. The Luau array carrying them is 1-based like any other, so indices[1] is the first triangle's first index, and vertex v has its x at positions[v * 3 + 1]. |
| Triangle winding | A triangle's front face is the one whose three vertices turn counter-clockwise as the viewer sees them. Put another way, cross(v1 - v0, v2 - v0) points out of the front face. |
| Back faces | A material culls its back faces by default, so a triangle turned away from the camera draws nothing where it stands and whatever is behind it shows through. |
| Normals | normals give the surface its outward direction and shade the face. Which side of a triangle draws comes from the index order alone, so a normal pointing away from the side the winding draws leaves that face standing and shades it as if the light were behind it. |
| UVs | (0, 0) samples the first texel of an image — its top-left corner — and v grows downward through it. |
| Axes | Model space carries the world's basis: +X right, +Y up, and -Z the direction an entity's transform.forward points. |
Winding is what the author states: the inward faces of a room and a sky shell
seen from inside are geometry meant to be wound the other way, and asset.create
writes the index order it is handed. Two levers turn a face around:
indices = { 0, 2, 1 } -- the same three vertices, reversed
asset.create("material", "two_sided", { -- or keep the order and draw both sides
shader = "@builtin::shaders.pbr", render = { cull = "none" },
})
cull = "front" draws the far side by itself, which is what an inverted-hull
outline is built on, and "back" is the default. .shader documents the whole
render block a material's mat.yaml carries.
A mesh states which way its own triangles face, so the question is answerable off the geometry — every triangle whose cross product agrees with the normal it carries is wound to draw the side that normal points at:
local cpu = meshRef:load()
local g = cpu:geometry()
local function facesOutward(t: number): boolean -- t counts triangles from 0
local i = { g.indices[t * 3 + 1], g.indices[t * 3 + 2], g.indices[t * 3 + 3] }
local function p(k: number, c: number): number return g.positions[i[k] * 3 + c] end
local ax, ay, az = p(2,1) - p(1,1), p(2,2) - p(1,2), p(2,3) - p(1,3)
local bx, by, bz = p(3,1) - p(1,1), p(3,2) - p(1,2), p(3,3) - p(1,3)
local cx, cy, cz = ay*bz - az*by, az*bx - ax*bz, ax*by - ay*bx
local n = i[1] * 3
return cx * g.normals[n+1] + cy * g.normals[n+2] + cz * g.normals[n+3] > 0
end
local inward = 0
for t = 0, #g.indices // 3 - 1 do
if not facesOutward(t) then inward += 1 end
end
cpu:unload()
Built-in primitive meshes
The engine ships blockout geometry every world references by bare name
({ model = "cube" }). Each one is centred on its local origin, so an entity's
scale reads directly as the size the mesh covers in the world:
| Mesh | Extent (x, y, z) | Local AABB |
|---|---|---|
cube | 1 x 1 x 1 | -0.5, -0.5, -0.5 .. 0.5, 0.5, 0.5 |
sphere | 1 x 1 x 1 | -0.5, -0.5, -0.5 .. 0.5, 0.5, 0.5 |
cylinder | 1 x 1 x 1 | -0.5, -0.5, -0.5 .. 0.5, 0.5, 0.5 |
cone | 1 x 1 x 1 | -0.5, -0.5, -0.5 .. 0.5, 0.5, 0.5 |
octahedron | 1 x 1 x 1 | -0.5, -0.5, -0.5 .. 0.5, 0.5, 0.5 |
rounded_box | 1 x 1 x 1 | -0.5, -0.5, -0.5 .. 0.5, 0.5, 0.5 |
capsule | 0.5 x 1 x 0.5 | -0.25, -0.5, -0.25 .. 0.25, 0.5, 0.25 |
torus | 1 x 0.3 x 1 | -0.5, -0.15, -0.5 .. 0.5, 0.15, 0.5 |
plane | 1 x 0 x 1 | -0.5, 0, -0.5 .. 0.5, 0, 0.5 |
ground_quad | 1 x 0.1 x 1 | -0.5, -0.05, -0.5 .. 0.5, 0.05, 0.5 |
ground_hex | 1.732 x 0.1 x 2 | -0.866, -0.05, -1 .. 0.866, 0.05, 1 |
shader_ball | 0.933 x 1 x 0.933 | -0.467, -0.5, -0.467 .. 0.467, 0.5, 0.467 |
plane is a flat quad in the XZ plane, so it has no vertical extent at all and
scaling it on Y moves nothing. meshRef:load():getBounds() reads the same
numbers back off any mesh, built-in or authored.
Put a mesh on an entity
A component draws a mesh through a resource("mesh") field, which accepts the mesh
by name, by ref, or as a MeshHandle — the component resolves whatever you give
it. The builtin Model and SkinnedModel both expose a model field:
entity.find("box").component.add("Model", { model = "tri" }) -- by name
entity.find("box").component.add("Model", { model = asset.ref("tri") }) -- by ref
entity.find("box").component.add("Model", { model = meshHandle }) -- a MeshHandle (below)
Runtime mesh — no asset
Build geometry straight onto the GPU and render it without creating an asset. The
resource("mesh") field takes the returned MeshHandle directly:
local h = renderer.mesh.create({ positions = {0,0,0, 1,0,0, 0,1,0}, indices = {0,1,2} })
entity.find("box").component.add("Model", { model = h })
Its geometry reads back off the handle. renderer.mesh.geometry(mesh) returns the
same flat-array MeshGeometry create takes, carrying every stream the mesh has —
so a runtime mesh can be inspected, edited, and rebuilt without ever becoming an
asset:
local geom = renderer.mesh.geometry(h) -- { positions, indices, tangents?, ... }
geom.positions[1] = 0.5
local h2 = renderer.mesh.create(geom)
An optional stream is present only when the mesh carries one, so geom.tangents
answers whether the mesh has a tangent basis and geom.uvs1 whether it has a
second UV set.
Read geometry — meshRef:load()
:load() decodes data.zmsh into the CPU store and returns a CPU handle carrying
vertexCount / indexCount:
| Method | Returns |
|---|---|
cpu:getVertices() | array of { pos = {x,y,z}, normal = {x,y,z}, uv = {u,v} }, one per vertex |
cpu:getTriangles() | array of { v1, v2, v3 }, each a full vertex (the shape above) |
cpu:getBounds() | { min = {x,y,z}, max = {x,y,z} } — the local-space AABB |
cpu:geometry() | the complete MeshGeometry — flat arrays, every stream the mesh carries |
cpu:encode() | the data.zmsh bytes |
cpu:unload() | drop this guid's CPU copy |
local cpu = meshRef:load()
local v = cpu:getVertices()[1] -- { pos = {x,y,z}, normal = {x,y,z}, uv = {u,v} }
local aabb = cpu:getBounds()
cpu:unload()
By default the CPU copy is transient (decode → read → unload). Set keepCpu to
keep it resident — see Settings.
Write geometry back — meshRef.primary
The inverse of reading: encode a MeshGeometry and write it to the asset's
primary content file. The assetType picks the write up and re-uploads the live
GPU mesh, so the change shows without a reload:
local geom = renderer.mesh.geometry(meshRef)
geom.tangents = computeTangents(geom) -- add a stream it did not carry
vfs.write(meshRef.primary, renderer.mesh.encode(geom), { overwrite = true })
meshRef.primary is the asset's data.zmsh path. This is the route for a
stream with no setting of its own; keepCpu + meshRef:setVertices moves
positions in place, and lightmapUvs regenerates the second UV set (both
below).
GPU handle — meshRef:handle()
:handle() uploads the geometry (Disk → CPU → GPU) and returns a MeshHandle,
cached on the asset so every consumer shares ONE GPU entry. A component does this
for you when you pass it the mesh; call it directly only for the runtime/low-level
path.
Settings — meshRef:settings() / meshRef:setSettings(patch)
A mesh's settings are declared in settings.luau and live in settings.yaml.
:settings() returns them with every declared default filled in.
:setSettings(patch) writes any subset — sibling settings are preserved, and
a patch that fails the schema is refused, naming the field and why.
| Setting | Type | Default | Meaning |
|---|---|---|---|
keepCpu | boolean | false | Keep the CPU geometry copy resident — for an in-use mesh — so geometry can be read repeatedly without re-loading. |
lightmapUvs | string | "keep" | The mesh's second (lightmap) UV set. "keep" leaves the stored geometry's set as-is; "generate" unwraps a fresh non-overlapping set; "none" strips it. See below. |
clusterLod | boolean | true | How deep the cluster set is built. Every static mesh draws through clusters, from the frame it materialises (its lod0 set) until its bake lands. true (default) builds the LOD hierarchy; false builds one level, drawn at full detail. |
local s = meshRef:settings() -- { keepCpu = false }
meshRef:setSettings({ keepCpu = true })
keepCpu is an active lifecycle control, and only ever holds geometry for a mesh
that is in use (GPU-resident):
- Setting it on while the mesh is in use loads the CPU copy immediately.
- Setting it on for a mesh that is not in use loads nothing — the copy is retained the next time the mesh is materialised, so an unused mesh is never pinned in memory.
- Setting it off drops a resident CPU copy (the renderer keeps drawing from the GPU).
clusterLod is the same kind of lifecycle control, for the depth of the
cluster set rather than the CPU copy:
- Setting it off on a mesh in use rebakes it as one level of clusters
(
single), drawn at full detail. - Setting it on on a mesh in use rebakes the LOD hierarchy (
dag). - A mesh that is not in use bakes at the depth the setting asks for when it next materialises.
Acting on either setting for a mesh in use can read the mesh's stored geometry,
and where the geometry is fetched on first read that read spans frames.
meshRef:settingsPending() answers whether one is still in flight; once it
answers false, every settings change made so far has reached the mesh.
Second UV set (lightmapUvs)
A mesh can carry a second, independent UV set (uv1) alongside its texture UVs
(uv) — a distinct per-vertex channel, non-overlapping where the texture UVs tile
and reuse charts. lightmapUvs controls whether the stored geometry keeps one,
generates one, or drops it:
| Value | Effect |
|---|---|
"keep" | Leave the stored geometry's uv1 as-is (the default). |
"generate" | Unwrap a fresh non-overlapping uv1 (near-planar charts packed into [0,1] with gutters). |
"none" | Strip the uv1 set. |
meshRef:setSettings({ lightmapUvs = "generate" })
Changing the setting reprocesses the asset's geometry in edit mode; a change in
play mode never rewrites the source. The reprocess round-trips the geometry through
the codec, so positions, normals, tangents, skinning, and the skeleton are preserved
— only the uv1 set changes. A runtime mesh carries uv1 the same way: pass uvs1
(or unwrapUvs = true) to renderer.mesh.create.
Read vertices — meshRef:getVertices()
With keepCpu on for an in-use mesh, read its vertices straight from the resident
CPU copy:
meshRef:setSettings({ keepCpu = true }) -- on an in-use mesh
local verts = meshRef:getVertices() -- { { pos = {x,y,z}, normal = {x,y,z}, uv = {u,v} }, ... }
If the CPU copy isn't resident — keepCpu is off, or the mesh isn't in use — it
errors loudly, naming the reason and the remedy, rather than silently loading a
throwaway copy:
mesh:getVertices: cannot read vertices — keepCpu is off, so no CPU copy is kept.
Turn on CPU retention for an in-use mesh first: meshRef:setSettings({ keepCpu = true }).
Morph targets — meshRef:morphTargets()
A mesh can carry morph targets: shapes it blends towards, each a per-vertex displacement from the base geometry. Every target carries the name it was authored under, and an imported model keeps the one its source file gave its blend shapes, so content addresses a shape as the thing it means rather than as the ordinal it happened to import at.
for i, name in meshRef:morphTargets() do print(i, name) end -- 1 Eyes_Blink
An entity blends them with ecs.MorphWeights, weight i driving target i.
renderer.mesh.morphWeights(mesh, { Eyes_Blink = 1 }) builds that ordered array
from weights named by shape — shapes left out sit at 0, and an unknown name
errors listing the ones the mesh carries.
Discovery
| Call | Shows |
|---|---|
asset.list("mesh") | every registered mesh |
asset.inspect("<name>") | a text report about the mesh (metadata, source, status) |
meshRef:settings() | the resolved settings |
meshRef:settingsPending() | whether a settings change is still reading the stored geometry |
meshRef:morphTargets() | the shapes it blends towards, named, in target order |
For the on-disk shape of a mesh asset, run asset.inspect("<name>") or read the
type's type.yaml — those are the live source of truth.
Related
renderer.mesh.*— the GPU/CPU resource layer (createa runtime GPU mesh,encode/decodetheZMSHpayload,readbackGPU → CPU).:load/:handlewrap it; seeman renderer..material+.shader— what shades the geometry..bundle— container assets that carry.meshinteriors.
In the Inspector
Selecting a mesh opens its Geometry section, after the asset's identity and before its Settings. It reports what data.zmsh holds, decoded once per payload version: vertices, triangles, the bounds and their size, the vertex streams beyond positions (normals, uvs, uvs1, colors, tangents, skinning), and, when the mesh has them, its skin joints and morph targets. While the renderer holds the mesh's cluster hierarchy, the section also reports its cluster and level counts. A payload is one index stream, so the section names no submeshes. A mesh folder with no data.zmsh says so.
Interface
What this asset declares: the schema it conforms to, what it exposes, and the rendered structured payload.
conforms to
zero/asset-type/v1⌬ Spec
suffix
.meshcontainernoprimary aliasesinit.luau, init.luaoptional files.metadata, data.zmsh, init.luau, init.lua, settings.yaml, preview.pngExposed API
⌬ Hooks
onChange(ref: any, change: { [string]: any }) → void
React to a write inside this mesh asset. A write to the stored geometry re-uploads it into the GPU mesh registered under this asset's guid, so everything already rendering it draws the new shape. A settings write — to the declared `settings.yaml` values file, or a legacy `.metadata` block for content published before it existed — acts on whichever fields it named: `keepCpu` and `clusterLod` are lifecycle settings governing PER-ENGINE state (a CPU copy in this process's store, a cluster set on this process's GPU), so they act on an already-materialised mesh right away for EVERY origin — a synced peer's engine holds its own CPU/GPU state that the write says just as much about as the one that made it; `lightmapUvs` reconciles the second (lightmap) UV set — "generate" (re)creates a non-overlapping unwrap into the second UV set, "none" strips it, "keep" leaves the stored geometry untouched. The re-cook round-trips the geometry through the codec, so tangents, skinning, and the skeleton are preserved; it is the one part of this hook that WRITES persisted content, so it alone stays local-origin-only (a peer's own re-cook of the same settings would just be a second, redundant write of the same bytes the sync already delivered) and play-mode-gated (the play-lock): a setting flip never rewrites the persisted geometry while the world is running; a resident mesh re-materialises on its next fetch. `path` the written file, `asset` the container folder, `kind` "edited"/"seeded", `origin` "local"/"remote", `settings` the validated values and `settingsChanged` what changed when `path` is the settings values file.
| arg | type | description |
|---|---|---|
| ref | any | The AssetRef<mesh> for the changed container. |
| change | { [string]: any } | `{ path, asset, kind, origin, settings, settingsChanged }` — |
Sub-parts
Everything contained inside this part. Assets are composite children (clickable cards). Files are leaf payloads. Expand any row to view its source.
Showing up to 3 levels deep. Open a part to see what it holds further down.
93+items
# assetType (asset type)
An asset type says what a kind of asset is: the files an asset of the kind is
made of, the code that gives it behaviour, the interface a person uses to work
with it, and how it is found, pictured and checked. A `<typename>.assetType/`
folder declares a type called `<typename>`, and every `<name>.<typename>/`
folder is an asset of it. `assetType` is itself a type, so a type is an asset
like any other: it is created, versioned, installed and contributed the same way.
## Making a type
`asset.create("assetType", "<typename>")` scaffolds a type. The type registers
the moment its folder is written, anywhere under `/zero/source/`, and from then
on `<name>.<typename>/` folders are its assets. An asset of a type that is not
registered yet is not an asset, so make the type first.
A type folder is a root: `~.x` in any file inside it names the type's own `x`,
wherever the type is installed. Keep the type's whole implementation inside its
folder and reach it that way.
What a type folder must hold is set by this type's own `type.yaml`, the same way
every type's `type.yaml` sets what its assets hold. `asset.validate` on a type
reports anything missing, and the scaffold produces a folder that passes.
## What sets your type's schema (`type.yaml`)
`type.yaml` is the structural contract every asset of the type is checked
against. The keys below are the ones the engine reads; a key not listed here is
ignored.
| Key | What it controls |
|---|---|
| `suffix` | The folder suffix of the type's assets. Defaults to `.<typename>`. |
| `primary` | The file or files that are the asset's content: what the asset resolves to when it is used. |
| `entry_files`, `primary_aliases` | Further files that load as the asset or share its identity. |
| `is_container` | Whether the suffix stays in the identities of what the asset holds. |
| `is_opaque_walker` | Whether a walk of the tree stops at the asset and registers only its primary. |
| `plural_dir`, `classify_exts` | Loose files in a folder of this name, or with these extensions, are assets of the type. |
| `is_text_default` | Whether the asset's files are text unless stated otherwise. |
| `broadcasts_subassets` | Whether assets nested inside an asset of the type are found by their own short name. |
| `root` | `true` makes every asset of the type a root, so `~.x` inside one names that asset's own `x`. |
| `contract_ref: { field, category }` | A key of the primary file that names another asset the asset is built on. |
| `files: { required, optional }` | The files an asset holds (below). |
| `allow_unlisted` | `false` reports every file the type does not declare. |
| `aliases_from: [{ file, key }]` | Further names an asset answers to, read from its own files. |
| `settings:` | Typed per-asset settings (below). |
| `inspector:` | The interface people use for the type and its assets (below). |
| `preview:` | How an asset is pictured (below). |
| `references:` | Names of other assets the type's files carry (below). |
| `indexing:` | What search reads for each asset (below). |
| `describe:` | How a missing description is generated (below). |
### The files an asset holds
Each entry under `files.required` or `files.optional` is
`{ path, kind: file | folder, one_of_group?, content? }`. `*` in a path matches
within one segment. Entries that share a `one_of_group` are satisfied by any one
of them. `content` lists checks on the file's bytes:
| Check | Passes when |
|---|---|
| `non_empty` | the file has content |
| `json_field { key, type?, non_empty?, accept? }` | the JSON holds the key (a dotted path) with that shape |
| `yaml_field { key, type?, non_empty?, accept? }` | the same for YAML |
| `described { keys?, headings?, message? }` | the file says what the asset is (see "Ask your assets to say what they are") |
Each check takes a `message`, which is what the author is told when it fails.
### Settings
`settings: { values, schema: { type | instance, export? } }` gives every asset
typed settings. `values` names the file an asset's values live in; `schema.type`
declares one schema for the whole type, and `schema.instance` names a file each
asset declares its own in. `asset.create` writes the resolved values, and a write
to the values file reaches `onChange` with what changed.
## The human side (`inspector.luau`)
A type is used by people through its Inspector, and without one the type is
usable only from code. The Inspector is the type's interface for a person: for a
procedural graph it is the graph editor, for a kit it is the tracking view of
what the kit holds and what is still missing. Writing it is part of writing the
type.
`inspector: { type, instance? }` in `type.yaml` names the files. `type` is the
interface for the type itself and for its assets; `instance` names a file an
asset may ship to add its own. Each returns:
```luau
return {
sections = { { id = "graph", title = "Graph", build = function(ctx) ... end } },
replace = { contents = { ... } }, -- replace a default section by id
order = { "graph", "contents" }, -- section order
}
```
`build(ctx)` receives the asset and builds the section. A section's `title`, and
the text its rows carry, is a string or an `i18n.text` value read from a catalog
(`i18n.text("inspectors.mesh.title")`), which follows the editor's language. `asset.create("assetType",
...)` scaffolds a working `inspector.luau` to start from. `asset.validate` on a
type that names no inspector, or names one that does not load, reports it.
## Say what your type is for (`about.yaml`)
Every type carries an `about.yaml` that says what the type is for, in three
plain-language texts:
```yaml
intent: "Sets the typeface of the text in your game: UI, labels, signs and 3D lettering."
function: "A font, from a TTF, OTF or WOFF file, made available to all of Zero's text, including 3D text."
usage: "You choose it by name for your UI or your text, and the text appears in that typeface."
```
- **`intent`** is what the type is for in someone's game, in the words a game maker uses. Up to 800 bytes.
- **`function`** is how it works today: what an asset of it is made of and what the engine does with it. Up to 1500 bytes.
- **`usage`** is what a person or an agent does with one. Up to 1200 bytes.
Write about the type as a whole, not one asset of it, and say what it does now.
Agents read these texts to learn when to reach for the type and how; people
searching find the type by them. The scaffold starts each text with
`[[describe]]` and a note on what to write; a text that still starts that way
counts as not written. `services describe` can write them.
They show up in `asset.inspect` as `detail.about`, in
`assetTypeRef:getAbout()`, at the end of the type's `types/<typename>` guide,
and at the top of the type's Inspector.
## Ask your assets to say what they are
A type can require each asset to carry a description: a file, written by the
author or generated, saying what the asset is and what it is for. Declare it with
a `described` check:
```yaml
files:
required:
- path: "README.md"
kind: "file"
content:
- { check: "described", message: "Say what this widget is for and where it goes." }
```
A file that is missing, empty or still has a line starting with `[[describe]]`
fails the check, and so does a YAML file whose named `keys` are missing or over
their `max_bytes`, or a Markdown file missing a named heading. `asset.validate`
names the asset, and a push that carries it is refused with `message`.
`services describe` finds such assets and writes their descriptions.
`describe:` steers what is generated:
```yaml
describe:
guidance: "Say what the widget shows, what input it reacts to, and where it fits in a UI."
examples:
- "<identity of an asset of this type whose description reads well>"
```
An example is an asset of your type whose description is a true, written one,
never a placeholder and never the asset being described; with none, leave
`examples` empty. A type whose assets the model cannot perceive (a sound, a
motion) says so in `guidance`, and the description is drawn only from what the
asset's own files state.
Start the file unwritten in your `template/`: a heading and one line starting
with `[[describe]]`, never example content, which every new asset would carry
verbatim. A type whose assets say what they are through their own content (a
mesh, a texture) declares no check.
## The type's code (`behavior.luau`)
`behavior.luau` returns one table. These are the keys the engine reads from it;
anything else on the table is the type's own.
### What the type's assets can do
| Key | Shape | What the engine does with it |
|---|---|---|
| `ref` | `{ [name] = function(self, ...) }` | Every function becomes a method on every asset of the type: `ref:name(...)`. |
| `modules` | `{ [name] = require(".x") }` | Code the type's assets reach with `asset.containing(__FILE__).modules.<name>`. |
| `events` | an events schema | Events an asset fires. Readers subscribe through `ref.events.<name>:connect(...)`; the type fires them. |
| `refShapes` | `{ [method] = function(self) -> (typeExpr, source?) }` | The result type of a `ref` method for one particular asset (below). |
| `namePattern` | a Lua pattern | The names `asset.create` accepts for the type's assets. Default `^[A-Za-z][A-Za-z0-9_]*$`. |
| `derive` | `{ output, module, compute, fingerprint, matches?, required?, timeout? }` | A file computed from the asset when it is committed; a `required` one holds the commit until it lands. |
Some `ref` names mean the same for every type, and the engine calls them:
- **`instantiate(self, target?, opts?) -> (root, idMap)`**: puts the asset in a scene. Defining it is what makes the type scene-instantiable. Its return is a contract (below).
- **`inspect(self) -> detail`**: what `asset.inspect` reports for the asset beyond the common record (below).
- **`preview(self, opts)`**: renders the asset's picture; it wins over the type's `preview:` recipe.
- **`previewMaterial(self) -> { material }`**: the material a `surface` preview paints.
- **`saveDefinition(self)`**: writes the asset's in-memory edits back to its files in edit mode.
Some names are answered for every asset before the type is asked, so a `ref`
entry of the same name is never reached: `getSource`, `getBytes`, `getText`,
`exists`, `deps`, `canInstantiate`, `canPreview`, `preview` (which calls yours
when you define one), `meta`, `runtime`, `events`, `typeRef`, `modules`, and the
residency flags `has_backing_asset`, `has_runtime_changes`, `cpu_resident`,
`gpu_resident`.
### When the engine calls the type
| Hook | Signature | When it runs |
|---|---|---|
| `onCreate` | `(name, opts) -> { [path] = contents }` | On `asset.create`. Returns the files of the new asset, laid over the type's `template/`. Annotate `opts`: its type is the schema `asset.create` checks the caller's arguments against. A `""` key writes a single loose file. |
| `onReplace` | `(dest, files) -> files` | On an `asset.create` over an asset already at `dest`, after `onCreate`, so what the author added can be carried into the replacement. |
| `onRegister` | `(self)` | Once per asset, when it first registers: when the world loads, on a live `asset.create`, and when an asset arrives later (installed, pulled, synced). This is what makes an asset take effect the moment it is written. |
| `onChange` | `(ref, change)` | On every write to a file inside an asset of the type. `change` is `{ path, asset, type, kind, origin }`, plus `settings` and `settingsChanged` for a write to the settings values file; `kind` is `edited` or `seeded`, `origin` is `local` or `remote`. Filter on `change.path`. Writes to one asset arrive in order, each after the hook for the one before has finished, and the hook may wait on what it reads. A write the hook makes inside its own asset arrives later like any other, so keep the hook convergent. |
| `onDependencyChange` | `(ref, change)` | When a module the asset's code requires is re-run, or a name the asset carries starts resolving. `change.kind` is `dependency`. |
| `onDelete` | `(ref)` | When an asset of the type leaves the world: deleted, or the world it is in is swapped away. |
| `validate` | `(ref) -> { { code, message, severity?, path? } }` | On `asset.validate`, after the `type.yaml` checks, for the type's own rules. `severity` defaults to `error`. Runs on every push, so a type's rules hold at publish. A hook that fails reports `validate.hook_failed`. |
### Settings hooks
| Hook | When it runs |
|---|---|
| `settingsDefaults(name, opts, asked) -> table` | Once per `asset.create`, for values the schema's defaults do not give. |
| `settingsCorrections(values, name, opts) -> (corrections?, reasons?)` | On every settings write. A correction to a field the writer named refuses the write, giving the reason. |
| `settingsDerivedDefaults(values, explicit, changed)` | On a settings write, for fields derived from others. |
| `normalizeSettings(values, stored?)` | On every read and write, to map an earlier spelling of a field to the current one. |
| `settingsHealed(values, stored)` | On a read, to repair stored values. |
| `settingsSchemaProblem(schema, ref) -> string?` | When an asset's own schema is read; a returned string refuses the schema. |
`creationInputs` (`{ [opt] = true | selectorOpt }`) names `asset.create` options
that are inputs to the type rather than settings.
### Reloading
An edit to `behavior.luau` reaches existing assets with no restart.
## How an asset is pictured (`preview:`)
| `kind` | The picture is | The type provides |
|---|---|---|
| `instantiate` | the asset placed in the preview studio through its `instantiate` | `ref.instantiate` |
| `image` | a file the asset holds, fitted to the frame | `source: { files }`, the candidate files in order |
| `surface` | the asset painted on a mesh | `mesh`, and `ref.previewMaterial` |
| `custom` | whatever `ref.preview` returns | `ref.preview` |
```yaml
preview:
kind: surface
mesh: sphere
```
`angle: [yaw, pitch]` turns the camera. The studio owns the light, the backdrop
and the framing, so an asset pictures the same in every world. A type with no
`preview:` that defines `instantiate` still answers `ref:preview()`.
A picture is rendered when something asks to view it (the editor showing the
asset, `ref:preview()`), when previews are generated on request
(`assets.generatePreviews`), and when the asset is staged. An edit alone
renders nothing. The picture records what it was made
from, so a stale one is told from a current one.
## The names your type's assets hold (`references:`)
Where your type's own format or API carries the name of another asset, declare
it here:
```yaml
references:
keys: # a top-level key of the type's files that holds a name, or a table of them
- { key: "shader", type: "shader" }
args: # an argument of a call, counting from 1
- { call: "g:node", position: 2, type: "procNode" }
fields: # a field of the options table a call takes
- { call: "fx.beam", field: "material", type: "material" }
```
`type` is the category the name resolves in. `asset.declareReferenceKey`,
`asset.declareReferenceArg` and `asset.declareReferenceField` declare the same
from a script.
## What search reads (`indexing:`)
`indexing:` is how assets are found. Search reaches an asset only through what
its type's `indexing:` names, and adds nothing of its own. That includes you: an
asset indexed badly is one you cannot find later, when you are making something
else and need it. A type whose assets are never looked for says `indexing: none`.
```yaml
indexing:
data: # text, read from the asset's files
- role: identity
extractor: verbatim
source: { file: "README.md" }
- role: interface
extractor: symbols
source: { files: ["init.luau"] }
- role: implementation
extractor: verbatim
source: { files: ["init.luau"] }
content: # the picture
- role: identity
modality: image
source: { file: "preview.png" }
facets: # what a query filters and re-ranks by
file: ".metadata"
rigged: { type: bool, from: { components: { contains: SkinnedModel } } }
roughness: { type: number, filter: range, from: { file: "mat.yaml", field: "floats.roughness" } }
```
**Roles.** Each `data` and `content` entry has a `role`, and a query matches in
two slots:
- **`identity`**: what an asset is. For anything you look at, the picture; for anything else, the text that says what it is.
- **`capability`**: what it does and how it is made. `interface` (the signatures or shape a caller uses) and `implementation` (the body) both count as capability, and a type with code usually declares both.
Ask what a person would recognise an asset of the type by, and what it does, and
point each role at the file that says it. For text, that is the asset's
description: the file your template starts with a `[[describe]]` block, which the
author writes or `services describe` fills in (see "Ask your assets to say what
they are").
**`data`.** `extractor: verbatim` embeds a file's text as it is; `symbols` embeds
the declared signatures of code. `source` is `{ file }` or `{ files: [...] }`.
Never index a file name (`source: { field: name }`): one
or two words carry nothing to match.
**`content`.** `modality: image` embeds the picture itself, so a text query
matches what the asset looks like. A type whose assets are looked at makes the
picture its identity.
**`facets`.** What a query narrows and re-ranks the matches by. `file` is embedded
whole, so keys nobody declared in it are still found. Each further key is
computed for every asset and states a value for every one, including `false`: a
file can imply "not rigged" but never say it, and a query separates rigged from
unrigged only because the unrigged ones say `rigged: false`. A key is
`{ type: bool | number | enum, filter?: range, from }`, where `from` is
`{ file, field }` (a field of one of the asset's files, dotted for nesting),
`{ metadata: key }`, or `{ components: { contains | any_outside } }` (what the
asset's entities carry).
## What blocks a push
A push checks every asset it carries against its type. These refuse it:
- a required file that is missing, a `content` check that fails, a `described` file not written yet;
- an error from the type's `validate`.
Unlisted files and warnings do not. A `described` file blocks only a push that
changes the asset. When a push is refused, the message names each asset and what
it lacks. `world.publishBlockers()` and `zm status` read the same list without
pushing.
## Typing a result from the asset (`refShapes`)
A `ref` method whose result depends on the asset (one entry per child folder, per
row of a config, per control a map declares) can only be annotated as
`{ [string]: Thing }`, which accepts any key. `refShapes` states the result for
each asset instead: a function returning a Luau type expression for that asset,
and the module whose types the expression uses. Read the asset; never run it.
```lua
-- inputMap.assetType/behavior.luau
M.refShapes = {
activate = function(self): (string, string)
local names = {}
for _, control in ipairs(M.ref.controls(self)) do
table.insert(names, control.name .. ": Handle")
end
if #names == 0 then return "", "" end
return "{ " .. table.concat(names, ", ") .. " }", "@builtin::modules.zinput.scheme"
end,
}
```
With it, `map.jump:onPressed(fn)` is checked and `map.noexisting` is reported.
A per-asset type applies where the call site names which asset: a component
field whose default names it, or an annotation `AssetRef<"inputMap", "<identity>">`.
`asset.resolve(name, category)` carries only the category, and a name computed
at runtime carries nothing, so calls on those keep the declared return. The type
belongs to the call's result: index it there, or bind what you need from it
there, and the names stay checked.
`require("@builtin::assetTypes.assetType.shared.refShapes").published()` shows
what was published: `categories` (each type's method surface) and `returns`
(per-asset results by identity, then method). An identity missing from
`returns` was never published. Results are recomputed when an asset is written;
at most 64 assets of one type are read.
## Putting an asset in a scene (`ref.instantiate`)
A caller writes one piece of code against every instantiable type, so what
`instantiate` returns is fixed:
```lua
local Instantiable = require("@builtin::assetTypes.assetType.shared.instantiable")
M.ref = {
instantiate = function(self, target, opts)
local root = Instantiable.root(self, target, opts)
-- compose the asset under `root`, now
return Instantiable.result(self, root, idMap)
end,
}
```
**In.** `target` is the entity the asset lands under, or onto for a hierarchy
type; with none, the type spawns a new root. `position`, `rotation` (three
numbers in degrees, or a quaternion), `scale`, `name` and `temporary` mean the
same for every type: `Instantiable.root` applies them to a root you make,
`Instantiable.place` to one you adopt. Document any options of your own in the
hook's `--!arg` docs.
**Out.** `(root, idMap)`. `root` is the composed root, live when you return:
build synchronously, never defer to a component's `awake`. `idMap` maps each
original id to the spawned one, so a component that composes the asset again on
every load can pass it back in and keep references into the composition. Every
call goes through `Instantiable.result`, which refuses a root that is not live
and turns a missing `idMap` into `{}`.
## Inspecting an asset (`ref.inspect`)
`asset.inspect(ref)` builds a record every asset shares (`identity`, `name`,
`guid`, `source`, `typeName`, `typeDefinitionPath`, `scope`, `origin`,
`description`, `tags`), then calls `ref.inspect(self)` for the type's own
`detail`:
```lua
M.ref = {
inspect = function(self)
-- read the asset's own files, never runtime state: inspect works on an asset
-- that has not been compiled, uploaded or registered
return { --[[ what makes this asset what it is ]] }
end,
}
```
Without it, the record is the shared part alone, and an agent inspecting the
asset learns nothing of what it is. A hook that raises leaves `detail` empty and
sets `record.warning`.
## Discovery
- `asset.list("assetType")`: every registered type.
- `asset.inspect("<typename>")`: the type, its README and its `about.yaml` texts.
- `asset.resolve("<typename>", "assetType")`: the type itself.
▲ 0↑ born
# Model
Loads and displays a 3D mesh on an entity. Disabling the component hides the mesh. Supports procedural meshes (`cube`, `sphere`, ...), model files, and remote URLs. Tint and outline are shader effects applied to the mesh.
**Material lives on this component.** There is no standalone `Material` component — a material only renders where there is a mesh to render it on, so the `material` field is part of Model (and SkinnedModel). Set it at add time or assign the field later; properties are registry-wide.
Public fields: `model`, `material` (`AssetRef<material>`), `tintR/G/B`, `tintBlend`, `outlineR/G/B`, `outlineIntensity`.
Methods: `:setTint(color, blend?)` (color: `{r, g, b}` array or `{r=, g=, b=}` map), `:clearTint()`, `:setOutline(color, intensity?)`, `:clearOutline()`, `:setMaterialProperty(prop, value)`, `:getMaterialProperty(prop)`, `:getMaterialPropertyNames()`.
```luau
entity(id).component.add("Model", { model = "cube", material = "gold" })
entity(id).component.get("Model").material = "checkerboard" -- swap material
entity(id).component.get("Model"):setTint({1, 0, 0}, 0.5)
entity(id).component.get("Model"):setMaterialProperty("roughness", 0.2)
```
**Tint is synced.** Every `tint*` field replicates, so `:setTint` is the colour every participant sees. A colour that changes every frame on many entities (a stress map, a heat map, a damage flash, per instance colour) belongs in an instance-data lane instead: wear `@builtin::materials.instance_color` (or `instance_color_unlit`) and write `renderer.instanceData.set(id, 2, r, g, b, coverage)`, which replicates nothing and costs about a twentieth of a tint write. `guides { path = "topics/rendering-instance-data" }` has the lane layout.
▲ 0↑ born
# scene_instantiable (module)
The instantiation contract's shared implementation. An asset type opts into
scene instantiation by defining `instantiate(self, target?, opts?)` on its
behaviour `ref` table; this module carries the parts that mean the same thing
for every type, so a caller writes ONE piece of code against all of them.
```lua
local root, idMap = ref:instantiate(target?, opts?)
```
## The contract
**IN — the base opts.** `target` is an owning entity ref: the instance lands
under (or, for a hierarchy type like a bundle, onto) that owner. With no
target the type spawns a fresh root. `position`, `rotation`, `scale`, `name`
and `temporary` place that root and mean the same for every type. `rotation`
takes three numbers as pitch/yaw/roll in DEGREES, or four as a quaternion. A
type may honour more opts of its own — `params` for a type built from declared
inputs, `idMap` / `diff` / `sourceTag` for the bundle override contract — and
says so in its own `instantiate` docs.
**OUT — the two returned values.** Every type returns the same pair:
- **`root`** — the composed root, as an `EntityRef`. Composition is
**synchronous**: the root and everything the type built under it are live
the moment the call returns, so a caller can parent to it, read its
components and hand it on in the same statement. There is no frame to wait
for and no callback.
- **`idMap`** — the `originalId -> runtimeId` map naming what the composition
spawned, `{}` for a type with no addressable children. **Never nil.** A
component that re-composes the asset on every load (`Asset`) keeps this map
and passes it back in, which is how a cross-entity reference into the
composition — `SkinnedModel.skeletonRoot` pointing at a bone — survives a
reload.
The return is enforced, not merely described: `AssetRef` dispatches every
`ref:instantiate(...)` through `result` on the way out, whether or not the
type called it. A type that returns something else fails at its own call
rather than handing its caller a nil root or a map that is sometimes absent.
## Exports
- `M.root(self, target?, opts?) -> root` — stand the root entity for a type
that spawns one: parented to `target`, born temporary when
`opts.temporary`, named `opts.name` else the asset's own name, placed.
- `M.place(root, opts?) -> root` — apply the placement opts to a root the type
already has (the adopt path: a bundle exploding onto its target, a
sceneModule reconciling under one).
- `M.result(self, root, idMap?) -> (root, idMap)` — return through the
contract. Checks `root` is a live entity ref, normalises a missing map to
`{}`, and errors naming the asset type when either is something else.
- `M.isOwned(opts?) -> boolean` — whether a component drives this call. The
`Asset` / `SceneModule` components tag their own calls with `sourceTag`;
they hold the reference and re-compose on every load. An untagged call came
straight from `ref:instantiate(...)` and has no such owner.
- `M.own(root, self, idMap?) -> root` — hand an already-composed root to an
`Asset` component pointing at `self`, so the reference and its map persist
and the composition is rebuilt on the next load. The component adopts the
live composition rather than building a second one.
- `M.check(value, constraint) -> ok, reason?` — the `sceneInstantiable`
field-constraint validator, also registered under that kind on load.
## The field constraint
`Field.instantiableRef` emits `constraint = { kind = "sceneInstantiable" }`,
and this module registers the validator for it on load. Given a written value:
1. `nil` passes — the field is optional (no asset assigned).
2. Reads the value's identity — a bare string, or a table's `__ref` / `guid` /
`identity` / `name`.
3. Resolves it (`asset.resolve(identity)` — any category) and asks the
resolved ref `ref:canInstantiate()`: true iff the asset's type defines an
`instantiate` method.
This is what lets `Field.instantiableRef` accept by CAPABILITY rather than a
hardcoded type list — a new scene-instantiable asset type is accepted the
moment it defines the hook, with no edit to the field or its consumers.
▲ 0↑ born
# proc
Code-authored, graph-backed procedural generation for the Zero engine.
A graph is one function of `g`: `proc.define(function(g) ... end)`. Nodes are called by
family and name (`g.<family>.<name>({ ... })`), inputs are declared with `g:input(name, default, opts)`,
the returned table is the graph's outputs, and `io.output` is how the graph stands in the world.
`builder.module/README.md` holds every rule of that form.
This module owns the whole procedural model: the graph builder + data model,
the operation registry, the type system, the lazy cached evaluator, deterministic scoped hashing, and the node-family op
implementations. Authored `.procGraph` graphs are persisted via
`procGraph.assetType`; the `procgen.preview` / `procgen.spawn` operations live in
`procgen.toolbox` and are invoked as tools: `tools.use("procgen", "preview", args)`
from a script or an agent's `tool_exec` script, or `zero procgen preview` in the
shell. Generated
meshes/materials/textures/bundles are persisted through their own `.assetType`.
An op family can also come from outside this package: a `.procPack` asset
registers its ops in the same registry, called as `g.op.<id>(...)` exactly like the ones
shipped here (`proc.registerPack(name, ops)` is what it registers through, and
what a test or a tool calls to hold a family for the life of one VM).
`terrain.*` is one such pack, shipped in `terrainSystem.package/procPacks/`
and covering a whole landscape; `guides { path =
"systems/terrainSystem/documentation/terrain" }` is its guide.
To find out what the registry holds, use the `procgen` toolbox: `searchNodes`
searches it and `describeNode` describes one node or op. `proc.search_ops(query)` and
`proc.schema(id)` are the same answers in-process, for a graph builder looking
something up while it runs.
▲ 0↑ born
# assetType runtime
The machinery that reads a `<typename>.assetType/behavior.luau` and turns its
declarations into live behaviour.
A `behavior.luau` is a table of declarations — `ref` methods, `modules` shared
code, an `events` schema, a `namePattern`, `refShapes`, and the `onCreate` /
`onRegister` / `onChange` / `onDelete` / `validate` / `settingsDefaults` /
`settingsCorrections` hooks. None of it runs itself. This is the code that
runs it, and it lives here because reading a type's declarations is the
**assetType type's own behaviour** — the same way `material.assetType` owns
what a `.material` does.
## Sub-modules
| Module | Reads | Responsibility |
|---|---|---|
| `ref.module` | `ref`, `modules` | Builds every `AssetRef` envelope, attaches the shared metatable, and dispatches per-type methods and shared modules onto it. Also owns edit-mode write-behind persistence. |
| `create.module` | `onCreate`, `namePattern`, `validate` | Backs `asset.create`: resolves the type, checks the name against its pattern, runs `onCreate` (or clones `template/`), and writes the instance. `validate.module` beside it checks a creation `opts` table against the type's `createSchema`. Before running `onCreate`, resolves the type's settings chain (below) and hands the result to the hook as `opts.settings`. |
| `settings.module` | `settings`, `settingsDefaults`, `settingsCorrections` | A type's declared settings schema and where its values live (`type.yaml`'s `settings:` block). `Settings.resolveForCreate` layers the schema's own defaults, the type's own `settingsDefaults(name, opts, asked)` hook (a DEFAULT the schema can't express — a name-derived convention, a fact read off a payload the caller handed it, a value following a field `asked` shows a higher layer naming; any later layer overrides it), the Preset Manager, an importer's settings, the caller's `settings` table, then the type's own `settingsCorrections(values, name, opts)` hook (a CROSS-FIELD CONSTRAINT the schema can't express, applied last over the fully merged values). A correction over a default/preset/importer value lands silently; a correction that would override what the caller's own `settings` table explicitly asked for refuses the creation instead, naming the field and why. Neither hook is ever handed a table to mutate; each returns what it wants changed (`settingsCorrections` may also return a `{ [field]: reason }` table for that refusal message). `Settings.cloneValues` deep-copies a values table — `create.module` hands `onCreate` a copy, never the table it stamps. `Settings.valuesOf` / `writeValues` read and write an asset's own values file. |
| `typeYaml.module` | `type.yaml` | The folder a type lives in and its parsed `type.yaml`, decoded once per content version of the file. `settings.module` and `inspector.module` read their blocks through it. `about` / `aboutAt` read a type's `about.yaml` (`intent`, `function`, `usage`), leaving out any key that is empty or still starts with `[[describe]]`; `aboutRows` and `aboutMarkdown` lay the texts out for the Inspector and the `types/<name>` guide. |
| `inspector.module` | `type.yaml`'s `inspector:` block | Which inspector a type declares for its assets (`inspector: { type: ... }` or `{ instance: ... }`), and the declaration that file returns — the sections it adds, the generic ones it replaces, and their order — or the problem that kept it from loading. |
| `onRegister.module` | `onRegister` | Fires an instance's once-only registration hook — on arrival, on world load, and for `@builtin` library instances. |
| `changeDispatch.module` | `onChange`, `onDelete`, `refShapes` | Routes a VFS source write or a folder delete to the enclosing typed asset's type hooks. |
| `refShapes.module` | `refShapes`, `ref` | Publishes what every `AssetRef<category>` answers to, so a member read on an asset-typed value is checked against that category's real surface. |
| `events.module` | `events` | The per-asset event runtime behind `ref.events`. |
| `instantiable.module` | `ref.instantiate` | The instantiation contract an asset type opts into: `root` stands the root entity, `place` applies the base placement opts to one the type adopted, and `result` shapes the `(root, idMap)` every `instantiate` returns. Also registers the `sceneInstantiable` field-constraint validator. |
| `declaration.module` | `type.yaml` | The type's decoded declaration, read once and held until a write inside an `.assetType` folder; what every question about a type as a whole reads. |
| `previewRecipe.module` | `type.yaml` `preview:` | How an instance's `preview.png` is produced: which of the four recipes (`instantiate`, `image`, `surface`, `custom`) the type declares, and what that recipe needs. |
## Reaching it
```luau
local rt = require("@builtin::assetTypes.assetType.shared")
rt.ref.loadTypeBehavior("material")
```
Fields resolve lazily. The prelude requires the sub-modules directly, in
dependency order (`ref`, then `changeDispatch`, `onRegister.install()`,
`refShapes.install()`) — a table that required all six on load would take that
ordering away from it.
▲ 0↑ born
# Component (asset type)
A component is a reusable behavior + data unit attached to an entity.
The component file declares public fields, methods, and lifecycle
callbacks; the engine instantiates one per attachment and drives its
hooks every frame. Components are the primary unit of gameplay code —
movement, AI, animations, UI panels, network sync, and per-entity state
all live here.
## When to use one
- You need per-entity state that survives across frames.
- You need per-entity behavior driven by engine lifecycle hooks
(`start`, `update`, `onDestroy`).
- You want to expose configurable fields to designers / `entity.spawn`
callers / scenes without code changes.
If you need a stateless pure-function library, use a `.module`. If you
need to generate content declaratively, use a `.service`; compute work
belongs in a `.computeShader`.
## Where it lives
- Source: `/zero/source/.../<Name>.component/`
- Identity: `<Name>` (the `.component` suffix strips from the
identity, but the folder retains the suffix on disk).
- Folder shape:
- `init.luau` (or `init.lua`) — declares the component. **Required.**
- `README.md` — instance-level documentation. **Required.**
- `.metadata` — agent-editable tags + free-form fields. **Required.**
The structural contract is enforced by `asset.validate` against this
type's `type.yaml`.
## How to create one
Always start with the scaffold — it guarantees the folder shape
matches `type.yaml` from the moment it's on disk:
```luau
asset.create("component", "<Name>")
-- Creates: /zero/source/<Name>.component/
-- init.luau (canonical declare + public block + hooks stubbed)
-- README.md (instance README template)
```
After scaffolding, edit `init.luau` to fill in the actual fields and
behaviour. The component registers as soon as the file is written —
hot-reload is automatic; no engine restart required.
## How it operates
1. **Registration.** Writing `init.luau` into a `.component/` folder
indexes the asset, mints a guid (in the `.meta` sidecar), and
surfaces the type to the engine's component registry. Same-frame
visible to `entity.component.add`.
2. **Attachment.** `entity(id).component.add(name, data)` (or the
batch form `tools.use("entityOps", "addComponents", targets,
{ [name] = data })`) attaches an instance.
The `data` table's fields override the component's declared
defaults; omitted fields take the declared default. The `data`
values are copied into `public.*` BEFORE `awake()` runs.
3. **Lifecycle.** The hooks run in this order:
- `awake()` — same frame as attach, immediately after the data
copy. Initialise state, set up native bridges, register with
other systems. Other components on the same entity may not have
awoken yet.
- `start()` — the next frame. Every component on the entity has
completed `awake()` by then, so cross-component setup is safe
here.
- `update(dt)` — same frame as `start()`, then every frame after,
for as long as the gameplay clock is running. `fixedUpdate(dt)`
runs once per physics step, straight after it, with the step's
`dt`, so a frame can run it several times or not at all;
`editorUpdate(dt)` runs
while the clock is held. Which of them ticks is decided by
`engine.mode` AND `engine.paused` together — see **Per-frame
hooks and the two axes** below, and `man modes`.
- `onDestroy()` — when the component is removed via
`component.remove` or when the owning entity is despawned.
Release resources, drop native ECS components.
For the full list (`onEnable`, `onDisable`, `fixedUpdate`,
`editorUpdate`, `onPropertyChanged`, `onAssetReload`,
`onModuleReload`,
`onBeforeSave`, …) and the timing details, **read `man components`
— that is the canonical lifecycle reference. Generic component
patterns (declare blocks, public fields, sync, ECS bindings,
typed methods) all live there.**
4. **Public surface.** Fields declared in `public` are inspectable,
editable in the inspector, and persist with the entity. Methods are
declared on that same table with a colon —
`typed function public:method(args)` — and every one of them takes
the component ref as `self`, so a caller invokes it with a colon
too: `entity(id).component.get(name):method(args)`. That is the one
calling convention a component ref answers to: the reflection
methods `ref:fields()` / `ref:getField(name)` /
`ref:setField(name, value)` take the same spelling. A function the
chunk declares at file scope (`function helper()`) is the
component's own, reachable from its hooks under its bare name;
`public` is what the ref carries.
5. **Settings and the Inspector card.** `type.yaml` names the component's
settings schema as `settings: { schema: { instance: "init.luau",
export: "public" } }`: the `public` table IS the schema, and its values
live on each entity the component is on, so `asset.schema(componentRef)`
answers the Field descriptors and no values file exists. On an entity
the Inspector shows the component as a card whose generic `settings`
section edits those fields as typed controls, each edit checked against
the schema and written to the live component as one undoable operation.
`type.yaml` also names `inspector: { instance: "inspector.luau" }`: a
component that ships an `inspector.luau` beside its `init.luau` shapes
its card — the file returns `{ sections, replace, order }` (see
`assetTypes/README.md`, "`inspector:`"), each section's `build(ctx)`
reading `ctx.component` (the live component) and `ctx.selection.id` (the
entity). `SceneModule.component/inspector.luau` replaces `settings` with
the bound module's declared inputs.
6. **Events.** An `events` block declares typed moments the component
fires from inside — `events = { onHit = Event({ dmg =
Field.number(0, NoSync) }) }`. The owner fires with
`events.onHit:fire({ dmg = 10 })`; outside code subscribes through
the proxy's subscribe-only facade,
`entity(id).component.get(name).events.onHit:connect(fn)`
(`:once` / `:wait` too). Pass `Sync` as the event's second argument
to replicate a fire to other peers. Events let one component react
to another's moments without polling. Full reference: **`man
components`**, Events section.
7. **Hot reload.** Editing `init.luau` reloads the component
definition; live instances are reconciled with the new declaration
in-place, and each crosses one lifecycle boundary: `onDestroy()`
ends the running life, then `awake()` and `start()` construct the
one that replaces it, and `update()` resumes on the next frame.
`public` and `private` values carry across; module-level `local`s
belong to the chunk, so the new life starts with them at their
initial values and its `awake`/`start` are what fill them in.
Whatever the old life handed elsewhere — a subscription, a shared
registration, a spawned entity — is released in `onDestroy`, the
call that pairs with the construction about to run. An edit that
leaves the source byte-identical crosses no boundary at all.
A field carries its value across when something stated it: a key of
the `data` table the instance was attached with (a scene record's
data is that table), or a write made since. Every other field takes
the edited source's declared default, so changing a default reaches
every instance that left the field to it. An instance that states
its own value for a field whose default the edit changed keeps that
value, and the reload logs a `[HotReload]` warning naming the field,
the entity, the value kept and both defaults.
## Discovery
- `asset.list("component")` — every registered component (engine +
world + libraries).
- `asset.inspect("<name>")` — identity, scope, public fields,
lifecycle methods, source path, and this type README. `detail`
carries the component's declared `fields`, `methods`, and `events`,
parsed from its source — `tools.use("assets","describe", "<name>")`
renders it as markdown, so a component's events are visible without
opening its source.
- `ls /zero/source/<Name>.component` lists the files the folder holds.
- `entity(id).component.list()` — components attached to an entity.
## Per-frame hooks and the two axes
`engine.mode` (`"edit"` / `"play"`) and `engine.paused` (the gameplay
clock) are independent, and the per-frame hooks read them separately:
- `update(dt)` / `fixedUpdate(dt)` tick while **the gameplay clock is
running** — whichever mode the engine is in.
- `editorUpdate(dt)` ticks while **the clock is held** — because the
mode is `edit`, or because play is paused — and also alongside
`update` whenever edit is running.
Which gives four states, and this table is the contract:
| `engine.mode` | `engine.paused` | hooks that tick |
|---|---|---|
| `edit` | `true` | `editorUpdate` |
| `edit` | `false` | `update`, `fixedUpdate`, `editorUpdate` |
| `play` | `false` | `update`, `fixedUpdate` |
| `play` | `true` | `editorUpdate` |
Entering edit sets `paused = true` and entering play sets it back to
`false`, so a session that never writes `engine.paused` only ever sees
rows 1 and 3 — which is where the shorthand "`update` is the play tick,
`editorUpdate` is the edit tick" comes from. It is the two default rows
of a four-row table, not the rule.
**Row 2 is the one that surprises.** With `engine.paused = false` in
edit — set by hand, or by a tool that advances the gameplay clock to
sample it, such as a `capture.collage` given a `duration` — `update` and `editorUpdate`
both fire, once each, every frame. A component that declares both and
routes them to one body therefore runs that body **twice per frame**:
```luau
local function drive(dt) ... end
function update(dt) drive(dt) end
function editorUpdate(dt) drive(dt) end -- twice per frame in row 2
```
Nothing reports the doubling; it reads as an animation running at twice
its intended speed. To get exactly one tick per frame in every row, ask
which axis you meant and answer it in the body:
```luau
function update(dt) drive(dt) end
function editorUpdate(dt) if engine.paused then drive(dt) end end
```
A scene `entrypoint.luau` declaring the same two hook names is dispatched
by the scene loader, not by this table: there `update` needs play AND
running, `editorUpdate` needs edit, and play-paused ticks neither. Read
`man scenes` before carrying a component's rule to an entrypoint.
## Authoring conventions
- Annotate public fields with `--!desc` so the inspector + LSP show
documentation alongside the field name.
- Declare lifecycle hooks (`start`, `update`, `onDestroy`) explicitly
rather than relying on engine defaults — readers can see at a glance
what runs when.
- Keep `update(dt)` cheap — it runs every frame per instance. Move
heavy work to `start`, to a `.computeShader`, or behind a job (`man jobs`).
- For multi-entity batching, prefer
`tools.use("entityOps", "addComponents", targets, components)` over
a loop of `entity(id).component.add` — it takes many targets and
many components in one call.
## Common pitfalls
- **Naming.** The component identity is the folder stem with
`.component` stripped (e.g. `<Spinner>.component/` → identity
`Spinner`). Don't repeat the suffix in the identity passed to
`component.add`.
- **`init.lua` vs `init.luau`.** Either is accepted (it's a
`one_of_group: "entrypoint"`), but `.luau` is canonical. Mixing both
in one folder is invalid.
- **`update` in edit mode.** It runs whenever the gameplay clock is
running, and entering edit stops that clock — so `update` is idle in
a freshly-entered edit session and ticks again the moment anything
sets `engine.paused = false`. Use `tools.use("wld", "play")` to test
update logic under the mode it ships in; `tools.use("wld", "edit")`
to return. See **Per-frame hooks and the two axes** above.
- **Renaming.** Renaming the folder changes the identity. References
to the component are recorded against the component itself, so
renaming or moving it does not break them.
## Related types
- `.module` — pure-function library, no per-entity state.
- `.service`: a declarative content generator, no per-entity state.
- `.preset` — a captured component configuration, replayable into
`component.add` without re-typing fields.
- `.bundle` — entity hierarchies that may carry component attachments
as part of the bundle.
## In the Inspector
Fields: its execution order, hooks, and how many methods and events it declares, then the fields its `public` table declares as read-only typed controls holding their defaults, with **Check** and **Open** for its entry file. Instances: a link to every entity carrying it. What its card shows on an entity comes from the component's own `inspector.luau`.
▲ 0↑ born
# Module (asset type)
A module is a reusable, stateless (or globally-stateful) Luau library
that other code pulls in with `require()`. Modules are the unit of
factored-out logic in a world: helpers, math libraries, format
parsers, registries, shared algorithms.
## When to use one
- You have logic you want to share across multiple components,
services, scenes, or tools.
- You have a registry, lookup table, or shared cache that callers all
read from / write to.
- You want a stable API surface (`M.fn`) other code can depend on.
If the logic is bound to one entity, use a `.component`. If you want a
named UI of fields rather than functions, use a `.preset`.
## Where it lives
- Source: `/zero/source/.../<name>.module/`
- Identity: the source path with `/` written as `.` and each segment's
type suffix dropped — `/zero/source/demo/beats.module` is
`demo.beats`, and `/zero/source/kit.package/beats.module` is
`kit.beats`, the container's name kept and its `.package` suffix
gone. `asset.create` returns this string, `asset.resolve(...).identity`
reads it back, and the `require` forms below write it as
`<identity>`.
- Folder shape:
- `init.luau` (or `init.lua`) — module body. **Required.** Returns
a table (conventionally named `M`).
- `README.md` — instance-level documentation. **Required.**
- `.metadata` — agent-editable tags + free-form fields. **Required.**
## How to create one
```luau
asset.create("module", "<name>")
-- Creates: /zero/source/<name>.module/
-- init.luau (canonical `local M = {} … return M`)
-- README.md (instance README template)
-- `folder` places it in a subfolder of /source, and that subfolder
-- leads the identity: `{ folder = "lib" }` produces `lib.<name>`.
-- `into` authors it inside a resolved container (a package, a toolbox),
-- whose own identity leads the same way:
asset.create("module", "<name>", { folder = "lib" })
asset.create("module", "<name>", { into = asset.resolve("myPkg.package") })
```
## How it operates
1. **Registration.** Writing `init.luau` into a `.module/` folder
indexes the asset and registers the module path with the resolver.
2. **Loading.** The first `require("<identity>")` runs
the file's body and caches the returned table. Subsequent requires
from anywhere return the **same table** — this is the
shared-state lever for "module-local" registries. The body runs in
an environment of the module's own: a name it declares without
`local` is readable by the module's own functions and by no other
script, so an export reaches its callers through the returned table
and nothing else. Requiring a component's folder runs its chunk the
same way, once, and its `update`, `start` and `public` stay in that
environment; a component runs on an entity through
`entity(id).component.add(name)`.
3. **Hot reload.** Editing the file re-runs the module body and folds
what it returns into the table already loaded, so that table keeps
its identity: a caller that captured it with `require` reads the new
source's functions and values through the reference it already
holds, and a later `require` returns that same table. A key the new
source drops goes away; a key something set at runtime stays. The
body's own `local`s start again from their initial values, so state
that must outlive an edit lives in `modules.state()`
(`guides { path: "core/scripting-and-tasks" }`).
4. **Identity resolution.** A module is reached by its `<identity>`,
and where the caller lives decides which spelling of it resolves.
For a module in a **world**, from a caller in that same world —
another module, a component, a scene entrypoint, or an `execute`
chunk:
- `require("<identity>")` — the identity on its own, resolved
against the caller's own root. It carries every folder segment,
so a module created with `{ folder = "demo" }` is
`require("demo.beats")`.
- `require("@root::<identity>")` — the same root-relative
resolution, written out.
- `require(".sibling")` — a module in the same folder, by its bare
name; `..name` steps up one folder, `...name` two.
Logs, stack traces and hot-reload notices name that same module
`@local.source.<identity>` — its key under the world source root.
The `<identity>` tail of that key is the form above, so a trace
reading `@local.source.demo.beats` is `require("demo.beats")`.
For a module in a **library** — content under
`/zero/source/libs/@<lib>/` — `@<lib>::` addresses that library's
root from anywhere: `require("@builtin::modules.transform")`.
Mounting a world as a library moves its files under such a root and
flips the caller root to `@<lib>::`, which is what keeps the
root-relative forms above resolving across the move.
## Discovery
- `asset.list("module")` — every registered module.
- `asset.inspect("<name>")` — public functions, source path, this
type README.
- `ls /zero/source/<name>.module` lists the files the folder holds.
## Authoring conventions
- Return a single `M` table from `init.luau`. Top-level statements
with side effects run on first require — useful for one-time
initialization, dangerous if they touch the engine before it's
ready.
- Write exported functions as `typed function`, not `function`: the
argument types are enforced at the call and the checker knows the
signature, so a caller's mistake is reported where they made it.
- Annotate exported functions with `--!desc` / `--!arg` / `--!return`
/ `--!example` so the LSP, `tools.list`, and `asset.inspect`
surface them. A function with no `--!desc` is reachable only by
someone already reading this file.
- Declare a VALUE with `Field.<kind>(default, mode, description)`, and
a table whose members are reached through a metatable with
`--!members <TypeName>` — the latter gives the surface its accepted
member set, its type, and its documentation from one table. See
`guides { path: "core/authoring" }` for all four rules and
`modules/api/engine/entity.module/members.module` for a worked
member table.
- Keep module state in module-local upvalues or in `modules.state()`.
A name the body declares without `local` lives in the module's
environment, which every run of the module shares: a function the
previous run left in place reads what the reload assigned. Module-locals start again from their
initial values with each hot reload; the table `modules.state()`
returns is the same table on both sides of every reload.
- Prefer focused modules over kitchen-sink modules. If two halves of a
module have no shared state, split them.
## Common pitfalls
- **Cyclic requires.** A `require` chain that loops will return the
partial table (the half built before the cycle was detected).
Design for one-way dependencies.
- **Engine-time side effects.** Don't `entity.spawn` at module
top-level — the engine may not be ready. Expose an `M.init` and
call it from a scene entrypoint.
- **Hot-reload + module-local cache.** A module that caches
expensive computation in upvalues loses that cache on reload —
fine for dev, but be aware in performance work.
- **`init.lua` vs `init.luau`.** Either works, `.luau` is canonical.
- **Writing one while play runs.** A `/source` write made while play is
running lands on the play shadow, and a guarded play-exit discards it
unless it is kept. One `.module` write shadows `init.luau`, the
`README.md` and the `.metadata` together.
`vfs.write(path, bytes, { durable = true })` skips the shadow, and
`vfs.promotePlayShadow` keeps what is already on it — the `core/vfs`
guide has the model and what each route costs.
## Related types
- `.component` — for entity-bound state + lifecycle hooks.
- `.service` — generates content you don't have yet (a mesh, sound, texture, …) via a metered provider.
- `.tool` — for a single agent-callable function with a YAML schema.
- `.package` — to group several modules + components + scenes into
one shippable folder.
## In the Inspector
Exports: what the module exports (name, kind, signature), parsed from its source without running it, with **Check** (the LSP's diagnostics for its entry file) and **Open**.
▲ 0↑ born
▲ 0↑ born
# proc.builder
What a graph builder hands the author, in the call form: `proc.define(function(g) ... end)`.
- `g.<name>(args)` adds a node of the node asset the name resolves to, by asset identity: the nodebox and the
node's folder name (`g.points.disc`), a path (`g["/zero/source/fx/ring.procNode"]`), `~` (the nearest package,
effect or world root: `g["~/ring"]`), a guid, or a short name (`g.disc`). A path, `~` name or identity ending in
`.procNode` or `.procGraph` names an asset of that kind alone; one without the suffix names either, so where a
node and a part share a name it states the suffix. A name that matches two nodes, and a name that matches none,
are errors that name the candidates. Each call is recorded in the graph by the node's
guid. Arguments are a table naming inputs or positional values in the node's input order.
- What a call answers: a node with one output answers that output's socket, a node with several answers a
record of sockets by output name (`local f = g.core.normals.face(geo); f.geometry, f.area`), a node with none
answers nothing. The record does not depend on declaration order. The record handed whole to an input that
takes one value is refused where it is handed, naming the outputs to read (`.value`, `.report`).
- A name is called this way only for a node written in the call form (`builder.callsByName(path)` says which). A
name no node answers to is refused naming the nearest names a call resolves.
- A node's own source reaches its package's modules and assets with `require("~.<name>")` and
`asset.resolve("~.<name>", type)`: `~` there is the nearest `.package` folder, wherever the node sits in it.
- `g.op.<name>(args)` (and `proc.op.<name>`) adds a node of an op, for library nodes built on the ops. An op that
is a graph built with `g:node` (the `procNodes/` of the procgen package, registered under their graph ids) is
refused at the call, naming its folder and the node written in this form that answers to the same name
(`'g.op.points.filter' is .../procNodes/pointsFilter.procNode, a graph built with g:node, ...; call
g.points.filter`). An id no op carries is refused naming the ops that start the same way.
- `g:input(name, default, opts)` is typed by the default: `Float`, `String`, `Bool`, `Vec3`, `Color`, `List<T>`,
`Record` with its fields. `opts` takes `min`, `max`, `options`, `description`, `optional`, `plain` and `type` for a
default that does not state its type or no default at all. An empty table default with a stated `type` is that type's
empty value (`g:input("entities", {}, { type = "List<Entity>" })`). A vector written as a plain table reads its
components by name whichever form it was written in, the reading an op's params take: `{ 0, 0, 0, 1 }` answers `.w`,
and a pair (`{ a, b }`, `{ x, y }` or `{ x, z }`) answers its second component as `.y` and as `.z`.
- An input declared `plain = true`, an input with `options`, and an input of type Bool or String is a plain value
while the graph is built: `g:input` answers the value itself, `if`, comparisons and loops over it are allowed, and
it passes into calls as a constant. `plain = true` is how a count, a vector, a record or a list that chooses the
graph's structure is declared (`g:input("storeys", 5, { type = "Int", plain = true })`, `g:input("edges",
{ { bays = 3 } }, { plain = true })`): the graph is built again for each value it is given, it is not animated, and
it takes no per-element field. Any other number and vector input is a socket, which changes without a build and
takes animated values and fields. A table value reaches the build as a frozen copy. The graph records the plain
values it was built with in `graph.plain`, and each such declaration in `graph.inputs[name].plain = true`.
- Sockets take `+ - * / % ^ // ..` and unary minus, `math.*` and `vector.*` (each a `math.<name>` node;
`vector.create` over sockets is `math.vec2`, `math.vec3` or `math.vec4` by its component count), `x.name`,
`x.name = v` and `x[name] = v` for a computed name (which rebind `x`), `x[class][name]`, `x[n]`, and
`for index, item in list do ... end`. A source loaded
through `builder.prepareSource` reads plain `math`, `vector` and `rand` over sockets unchanged; `define` loads a
node or graph entry that way when it names them.
- `vector.create` has the width of its component count whether its arguments are sockets or numbers: two numbers
answer the Vec2 value `{ x, y }` and four the Vec4 value `{ x, y, z, w }` (what `math.vec2` and `math.vec4` make
of numbers), and three a Luau vector. `orient = vector.create(0, 0, 0, 1)` writes a Vec4 column.
- `g.compare(a, op, b)`, `g.where(cond, a, b)`, `g.pick(choice, alternatives)`, `g:fold(n, init, fn, opts)`,
`g:state(name, columns)`, `g:step(dt, substeps, fn)`, `g:check(cond, message)`, `g.has(geometry, class, name)`
and `g.copy(socket)` record the constructs the evaluator knows.
- `g:state(name, columns)` states the columns a step keeps, `{ [name] = { type, count } }`, as values the plan knows:
literals and inputs, records and lists of them, and what constructs and a `pure` `io.luau` call make from them. A
pure call's answer stands even when its arguments are cooked (a call reading the attribute names of a source's
points), so a `for` over its answer may add a column per name (`schema["s_" .. name] = specs[name]` over
`specs.keys`), and a step body writes them by the same names.
- `g:check(cond, message)`: a `cond` known while the graph is built refuses the build with the message (a true one
records nothing); a `cond` that reads a socket is recorded for the plan or the cook to answer. The build of a
node's own defaults, which is what indexes the node, holds such a refusal instead of raising it (`graph.refusals`),
and `forInputs` states it where the node is called or cooked as it stands, so a check over a plain input fires when
the call gives it the refused value.
- A plain Luau table filled inside a `for` body over sockets keeps one element (`rings[#rings + 1] = v`,
`table.insert(rings, v)`): the body is one pass for every item. `builder.check` refuses it with the code
`socket-loop-append`, naming the form to write: a local declared before the loop and assigned inside it, read after the
loop, where it is the list of every pass's value.
- `g.copy(value)` of a record or a list written in the graph (`g.copy({ total = 0 })`) is a value of its own that
member writes rebind, so a `for` body that fills it carries it from pass to pass.
- A class view made with a socket key (`names[i]`) and read as a value is a node of the scope it was made in, whichever
loop body first reads it.
- A `for` body is one pass that stands for every item. `index` is the item's place (a socket, from 1). A value built
outside the body that the body reads or assigns (`geometry[class][name] = v`, `record.name = v`,
`record[name] = v`) is carried: each pass starts from what the pass before left, the first from the value as it
stood before the loop, and after the loop the value is what the last pass left. The loop records the carries in
`scope.carries[id] = { init, result }` and the refs `loop_index`, `loop_carry` (the value at the start of a pass)
and `loop_result` (the value after the loop). A value the body only reads reads as it was. The emitter writes a
carrying loop as it was authored: a local that starts as `g:copy(value)`, the body's assignments to it, and the loop
reading that local after it (`proc.emit`).
- `g.clock.placement` (16 numbers, column after column), `g.clock.inverse`, `g.clock.rotation` (`{ x, y, z, w }`)
and `g.clock.entity` answer where the graph stands in the scene it is cooked in: each is a `construct.scene` node
that reads the scene the cook is handed (a graph standing on no entity has the identity placement and no
entity). `g.transform.point(p)`, `g.transform.direction(d)` and `g.transform.rotation(q)` carry a world position,
direction or quaternion into the graph's space through the inverse placement; each takes plain values, sockets or
Fields.
- `g:fold(n, init, fn, { stop = "changed" })` or `{ stop = function(before, after) return <Bool> end }`; `fn` returns
the state only. The fold runs `n` passes and ends early at the first pass whose stop holds.
- `g.element` is the element a Field evaluates over: `g.element.id` (Int), `g.element.P` (the position leaf that
`fieldc.withPosition` substitutes, also written `g.element.point.P`) and `g.element.<class>.<column>` for a column
of the domain the field stands in. Each read is a socket, so `g.element.id * 1.0` is a math node.
- `rand.unit(seed, ...scope, id)` (a global in a node source, also `builder.rand.unit`) is the Float field of unit
draws keyed by the seed, the scope (`graphId, nodeId, channel`, or `graphId, compositePath, nodeId, channel`) and
the element `id`, last (`g.element.id`). It adds an `io.luau` node
`{ module = "@builtin::systems.procgen.proc.fieldc.leaves", fn = "rand", args = { seed, { ...scope }, id },
returns = { "Any" } }` and answers its `value`.
- `geometry.names[class]` lists the attribute names a class holds (a list socket the plan answers);
`geometry[class].count = n` assigns the domain count, so `count` is no column's name (a value per element written
under it is refused); `geometry.material = m` assigns the material. `names` is
read by that name on any socket: what is read off it by a class (`geometry.names.point`) lists the names, and any
other read (`record.names[2]`, `record.names`) is the member a record holds under that name, as `count` and `keys`
are. `material` is assigned by that name on any socket.
- A member, an item, a column or the names of a value built outside a fold, step or stop body that the body reads
first is a node of the scope the value was built in: the body and every later fold read the one value, which the plan
writes once. A read keyed by a socket of the body (a fold's index) stays in the body.
- `g.clock.time`, `.frame`, `.delta` and `.duration` are sockets; `.fps`, `.frames`, `.loop` and `.realTime` are the
timeline's values.
- Luau's `< <= > >= == ~=` over a socket, and `and`, `or`, `not`, `if` and `while` that test one, have no value
while the graph is built. `builder.check(source)` names each one with the `g.*` form to write, and `define` holds a
node or graph entry to it. `and` and `or` test their left operand alone, so `plain or socket` passes. The check
follows names block by block as Luau scopes them: a local, a loop variable or a function parameter is its block's
own, and a parameter of a function that is not a fold, step or stop body is not known to be a socket (a comparison
over one is refused when the graph is built). Plain inputs are not sockets to the check. A plain Luau table indexed by a socket (`LANES[values.type]`) answers nil
while the graph is built, so the check refuses it with the code `socket-table-index`, which names `g.pick`.
- A list of values holds a nil as a value: `{ a, nil, b }` is three, and the second is a constant of nothing.
## Building a graph for other plain values
A graph whose plain inputs change structure is built again for the values a cook was given.
- `builder.forInputs(graph, inputs)` answers `graph` while its plain inputs hold what the build used, else the graph
built again from `graph.source` with the values `inputs` states (an input it leaves out takes its default); values
are compared by content, a record or a list included, and the result is cached by the source text and the values,
so a node written again is built again. `builder.KEPT_BUILDS_PER_SOURCE` of a source's builds are kept, the ones
used last. A build for other values keeps what the cook already made: only the nodes the new values change cook.
A plain input given an animated value, a per-element field or a value held on the device is refused by name. It answers `nil, reason` when the source cannot be built again or a value
is outside the input's options.
- An optional input with no default (`g:input(name, nil, { optional = true })`) is nil while the graph is built for a
call that leaves it unwired, and its socket otherwise, so `if input == nil then ... end` in a node source is decided
by what the call wired. The declaration states `wiring = true` and the graph records the wiring it was built for in
`graph.wiring[name]`; `forInputs` builds the graph again for a cook given no value for such an input (`values[name]
== false` states none in `rebuild` and `withInputs`).
- `builder.rebuild(graph, values)` builds again unconditionally; `builder.withInputs(values, fn)` runs a function
that calls `proc.define` and gives that graph the values. A node asset called with a plain value records it as a
constant param; the evaluator reads a node's graph through `forInputs` with those params.
## What a build resolves once
A build resolves each name once. The first call of a name `g.<name>(...)` in a build finds the node asset the name
answers to; the build keeps the asset, its inputs and outputs, and the envelope its calls record, and every later call
of that name in the same build takes them. Each `building` frame (a graph source, or a node source loaded by a call)
holds its own, by the graph's source and the name, on the thread that runs the build; the memo ends with the frame, so a
node written between two builds is resolved again by the next, and builds on other threads share nothing. The `Info` of
a node asset is kept by the node graph it was read from, and the line of each call is kept as a number that `siteOf`
states as `path:line` where it is read. `builder.setResolveMemo(false)` switches all three off for the builds after
(the answer is the setting before), and `builder.resolveCounts()` answers the calls that resolved a name (`resolved`)
and the calls a kept answer served (`hits`).
## What a build is made against
A graph source that has a base is built against it. The base of an entry source is the graph the last
`proc.takeHandOff` of its folder gave a Generator (`builder.setBase(graph)`, `builder.baseOf(entry)`), held weakly.
When the run of a build ends, a pass compares every node the run made with the base's node of the same id (`compareNode`
of `proc.eval.replan`, and the node's `op` field by field). A node that is equal is replaced in `graph.nodes` by the
base's own table, which is frozen, along with the tables under its inputs and params; a node that differs, or that the
base does not hold, is frozen where it is. The scopes are compared and the ones that differ are listed; they stay the
build's own tables. The graph states the result in `_diff`: `changed` (the ids of the nodes that differ or are new),
`removed` (the ids of the base's nodes the build did not make), `scopes` (the ids of the scopes that differ), `added`,
the counts of the nodes and scopes, `against` (the base, held weakly) and the edit counts of both graphs when the pass
ran. The planner (`replan.dirty`) reads it while the graph and the base stand as they did then, and compares only the
nodes it lists.
- A node that keeps the base's table keeps the place the base gave it in the order the nodes were made in. When this
build made it elsewhere, the graph states the new place in `_moved`; `builder.seqOf(graph, id)` answers the place a
node holds, and `graph:toJSON()` states the nodes with it (`builder.documentNode(graph, id, node)` is one node as the
document states it), so a document and an emitted source are those of a build that made every node itself.
- A table two graphs share is frozen, so a write into one raises. `proc.edit` replaces a node it writes into by a copy
of its own (the node, its `inputs` and its `params`) before it writes, and `builder.againstCounts()` counts those
copies beside the passes run (`passes`), the nodes filed from the base (`reused`), the nodes listed (`changed`) and the time the passes took (`ms`).
- `builder.setBuildAgainstBase(false)` switches the pass off for the builds after (the answer is the setting before).
A build of a source with no base, a build for other plain values (`forInputs`, `rebuild`) and a build of a node asset
are the graph their run made.
## What a build reads
A node or graph source is compiled with `builder.compile(src, "@" .. entry)` and run as a build with
`builder.building(entry, fn, ...)`. While a build runs, the source sees a `vfs` whose `read`, `readAsync` and
`readBounded` note each file for that build, and `builder.read` (`proc.read`) notes a file read from a module the
build calls into. When the build ends, the files it read are recorded against its entry source, replacing what its
earlier builds read: `builder.readsOf(entry)` lists them and `builder.readersOf(file)` the entry sources whose last
build read a file. A write of such a file is a `proc.sourceChanged` of each of those sources. `builder.forget(folder)`
drops the source text and the built graph the builder holds for a node or graph, so the next build that reaches it
runs its source again. The modules a build requires are recorded by the engine, which tells the node or graph type
through `onDependencyChange` when one of them re-runs.
## The wiring of each run of `graph.module`
`graph.module` binds its wiring into the builder (`builder.bind`) each time its body runs, with the metatable of the
graphs that run makes as `helpers.graph`, and the builder keeps that wiring on the metatable. A graph is wired by the
run that made it: `builder.run` declares its outputs, a value is wired into its refs, and `rebuild` builds an inline
graph again, each through the wiring of the graph's own run. The body runs again whenever the require cache lets go of
the module while a module that took its functions keeps them (`proc.define` is `graph.module`'s `define` as `proc`
loaded it), as a reset of the scripting state to a mark taken before the module loaded does; the graphs of both runs
are then built side by side.
## What a graph records
A call-form graph states `surface = "call"`. Each node it makes carries `call = true`, `seq` (the order it was made
in, which the emitter writes it back in) and `scope`. A node asset is `op = { __ref, type = "procNode", name }`; an
op is its id string; a construct is `op = "construct.<kind>"` with `construct = <kind>`. `graph.scopes` holds each
`for`, `fold`, `stop` and `step` body, `graph.states` each `g:state`, `graph.plain` the plain input values.
Each call and scope also records the line of the graph's source it was written on, `path:line`
(`builder.siteOf(graph, id)`), in a table the graph's document leaves out, so where a call sits moves no hash. The
plan carries it onto each call it writes out (`plan.flat.origins[<flat id>].site`, the graph's own calls in
`plan.flat.sites`), and a cook's failure is read at it: `describeFailure(flat, err)` of the eval module
(`@builtin::systems.procgen.proc.eval`) answers
`<path>:<line>: g.<call> failed: <why>`, the line being in the deepest source the person can edit.
The `inputs` of a construct node are refs (a constant is `{ kind = "const", value }`); `params` are constants.
| construct | inputs | params | answers |
|---|---|---|---|
| `member` | `value`, `name` | | the member of `value` |
| `member_set` | `value`, `name`, `to` | | `value` with the member written |
| `item` | `list`, `index` | | the item |
| `column` | `geometry`, `class`, `name` | | the column read (a read of `count` is the domain count) |
| `column_set` | `geometry`, `class`, `name`, `value` | | the geometry with the column written |
| `column_remove` | `geometry`, `class`, `name` | | the geometry without the column |
| `count_set` | `geometry`, `class`, `count` | | the geometry with that class's domain count set to `count`; a value per element is refused (no column is named `count`) |
| `material_set` | `geometry`, `material` | | the geometry with its material set |
| `names` | `geometry`, `class` | | the List of attribute names the class holds, answered at plan |
| `has` | `geometry`, `class`, `name` | | a Bool answered at plan: whether the class holds the column |
| `check` | `cond` | `message` | nothing; a `cond` that is false when cooked is an error stating `message` |
| `compare` | `a`, `b` | `op` (`eq neq lt le gt ge`) | a Bool |
| `where` | `cond`, `a`, `b` | | `a` where `cond` holds, else `b` |
| `pick` | `choice`, `alternatives` | | the alternative the choice names: by option name, by a list position (a false choice the first, a true one the second), or by `false` and `true` keys |
| `step_sample` | `step`, `value` | | the value of an animated `value` at the step's time |
| `step_between` | `step`, `index` | | the fraction (0 to 1) of the step's movement that source `index` of the step has covered; the evaluator defines it |
A call that has several outputs names them in refs: `{ kind = "node_output", node, output = "<name>" }`; a node with
one output has `output = nil`.
### Refs
The v1 refs (`const`, `graph_input`, `node_output`, `list`, `record`, `animated`) and:
| ref | stands for |
|---|---|
| `loop_item { loop }`, `loop_collect { loop, ref }` | the item of a `for` scope, and the list of `ref` over every item |
| `fold_value { scope }`, `fold_index { scope }`, `fold_result { scope }` | the value and the index (from 1) of a fold pass, and the value after the last pass |
| `fold_before { scope }`, `fold_after { scope }` | the state before and after a pass, inside a fold's `stop` function; `scope` is the `stop` scope |
| `step_state { scope }`, `step_index { scope }`, `step_result { scope }` | the state record, the step record (`dt time graphTime index first start`, with `step_sample` and `step_between`), and the state the step reached |
| `state { name }`, `clock { field }`, `engine` | declared state, the clock's field, the engine API namespace |
| `element { field, class? }` | the element a Field evaluates over: `field = "id"` its index, `field = "P"` its position leaf (no class), else the column `field` of the domain `class` |
### Scopes
`graph.scopes[id]` is `{ kind, parent, seq, ... }`:
- `for`: `list`, `item`.
- `fold`: `count`, `init`, `result`, and `stop` when it states one: `"changed"` (the pass ends the fold when the
state it produced equals the one it read) or `{ scope = <id of a stop scope> }`.
- `stop`: a scope whose `parent` is its fold. Its nodes read `fold_before` and `fold_after`; `result` is the Bool
ref the fold stops on. It is evaluated after each pass.
- `step`: `dt`, `substeps`, `result`.
The `plain` table of a graph and the `plain = true` flag of an input declaration state which inputs are build-time
values.
Part of `@builtin::systems.procgen.proc`.
▲ 0↑ born
▲ 0↑ born
▲ 0↑ born
# proc.demos
Procedural-generation demo recipes — graph builders that showcase the proc
system end to end. Part of `@builtin::systems.procgen.proc`.
- **`cashPile(proc, opts?)`** — scatters a billboard source over a disc of
points and merges it into a single mesh (the scatter → merge showcase).
▲ 0↑ born
▲ 0↑ born
▲ 0↑ born
▲ 0↑ born
▲ 0↑ born
# proc.eval
Lazy topological evaluator of a graph target, with per-eval memoization and cycle detection.
Part of `@builtin::systems.procgen.proc`.
## Graphs built in the call form
A graph built with `proc.define(function(g) ... end)` has `surface = "call"`. It records nodes (ops and
node assets), scopes (`for`, `fold`, `step`), declared state and the refs between them. It has no output
the evaluator is asked for: every node nothing reads is cooked from.
```lua
local play = eval.play(graph, { inputs = { count = 8 } })
local report = play:present({ time = 0, frame = 0, fps = 30, duration = 0 })
play:release()
```
`eval.play(graph, opts)` plans the graph, cooks from every node nothing reads and keeps what those nodes
return. `play:present(clock)` cooks at a clock (`{ time, frame, fps, duration }`; a `duration` above 0 is a
timeline with an end, where a seek replays from the start), `play:setInputs(inputs)` gives the graph other
inputs, `play:handle(id)` answers what a node cooked from answered last, `play:report()` says what happened
and `play:release()` lets go of everything the play holds, the Handles its output nodes answered included.
What a play stands in the world goes over to a present all at once. A present opens a stage (`eval.openStage`):
every show its cooks make (an `io.output`'s `show`) is prepared and set aside, and once the present's pulls are
cooked the stage commits (`eval.commitStage`) and every show stands in that frame. A value that stood and that the
present supersedes (a node's Handle cooked again, an output node no plan holds any more) is set aside as well
(`opts.standingRetired`) and let go of right after the commit, each release on a task of its own
(`eval.releaseStanding`), so the world never holds part of one present beside part of another. A present that raises
stands none of its shows, drops the nodes that made them (`eval.forgetValues`), and leaves what stood standing for the
next present to replace. A value that stands is never kept per key on a timeline with an end: a key it held before is
cooked again, and its show made again, when it comes back. A present cooks with the inputs the play holds when it
starts: inputs `setInputs` gives it while it runs are the next present's.
Every call-form graph cooks through the plan, whoever asks: `proc.eval(graph, target, opts)` of a graph with
`surface = "call"` plans it (`opts.inputs`, `opts.clock`), cooks what the plan waits on and then the target through
the flat graph, and a node asset used as an op cooks its graph the same way. `output:<name>` names an output of the
graph (`flat.outputs` holds each one lowered), and the nodes the eval cooked into its own cache are released apart
from what the answer reaches. Each node cooks at `opts.clock`: an animated value it reads, handed straight to an op
or through a node asset's input into a field op (`fieldc.pcallSampling`), holds its value at that time; one that
becomes a field stays a live field, rewritten for a later time without a cook.
`proc.historyKey` and `proc.liveDigest` of a call-form graph are queries of the same plan: the plan is made, read
and keyed exactly as a cook makes, reads and keys it (what it waits on is cooked, its leaves are keyed), and only the
final read of the flat graph answers the key or the digest. So a key asked before an eval and after it is the same,
and a plan a query made is kept keyed for the cooks after it.
`eval.plan(graph, opts)` is the plan alone, and `eval.references(graph)` the assets and ops the graph's calls
reach, by guid and id, without cooking.
### The document
```
graph.surface "call"
graph.nodes[id] { op, inputs = { socket = ref }, params = { name = ref | value }, scope = scopeId? }
graph.scopes { [id] = { kind = "for", list = ref } | { kind = "fold", count, init, result }
| { kind = "step", dt, substeps, result }, parent = scopeId? }
graph.states { [name] = { columns = ref } }
graph.inputs { [name] = { type, default, ... } }
graph.outputs { [name] = { ref } } -- what a graph used as a node gives
```
`op` is an op id (`"device.kernel"`, or `"op.device.kernel"`), a node asset
(`{ __ref = guid, type = "procNode", name }`) or `"construct.<kind>"`.
### The graph's hash
`proc.depHash(graph)` (`eval.depHash`) is the hash of two things: the document without its nodes (id, version,
metadata, timeline, inputs, outputs, composites, scopes, states, surface, plain values; `graph.documentHeader`) and the
sum of the nodes' digests, mod 2^52. A node's digest is the hash of its id and of the node as the document states it
(`builder.documentNode`: the node, with the place a build gave it when that is not the one its table holds). The sum
does not depend on the order of the nodes, so two documents of the same nodes hash equal however their nodes are held,
and a document read back from JSON hashes as the graph it was written from.
A graph keeps its sum with the count of edits it was stated at (`eval.depSum`). A graph built against its base
(`_diff`, see the builder README) states it as the base's sum less the digests of the base's nodes that the build
changed, removed or moved, plus the digests of those nodes in the new graph; the nodes it filed from the base are not
read. A graph with no base, one edited since the build, and a document state it from every node (`eval.depHashWhole`
is that sum, the definition the patched sum equals). `eval.depHashCounts()` answers the sums patched (`patched`), worked
out from every node (`whole`) and found kept (`held`), and the digests made. `eval.setDepHashFromNodes(false)` returns
to the hash of the serialized document folded in one pass; the two give different numbers for the same graph, so a key
built from a hash (a composite's version, the disk cache) misses once when the setting moves.
The refs are the v1 refs (`const`, `graph_input`, `node_output`, `list`, `record`, `animated`) and:
| ref | stands for |
|---|---|
| `loop_item { loop }` | the item of the `for` scope in the copy being written out |
| `loop_collect { loop, ref }` | the list of `ref` in every copy of the `for` scope |
| `loop_index { loop }` | the place (from 1) of the item in the copy being written out |
| `loop_carry { loop, carry }` | a value built outside the `for` scope as a pass begins: the value before the loop for the
first copy, what the copy before left for every other |
| `loop_result { loop, carry }` | the same value after the last copy |
| `fold_value { scope }`, `fold_index { scope }` | the value and the index (from 1) of the pass of a fold |
| `fold_result { scope }` | the value after the last pass |
| `step_state { scope }`, `step_index { scope }` | the state record and the `step` record of a step body |
| `step_result { scope }`, `state { name }` | the state the step has reached |
| `clock { field }` | `time`, `frame`, `fps`, `duration` or `delta` of the graph's clock |
| `engine` | the engine API namespace |
### The plan
- A node asset built in the call form is written out where it is called, its inputs bound to what the
call was given. A call nothing reads is cooked from through every output it declares.
- A `for` body is written out once per item of its list, a `fold` body n times in a chain, a `pick`
takes the one alternative its choice names. A copy is keyed by its content, so reordering, inserting
or removing items cooks only what is new.
- A count or a list a scope needs is a plan value: a constant, an input, or the answer of a node, which
the plan cooks and lists in `report.waits`. A column left on the device is not a plan value; reading it
back with `transfer.readback` is, and the wait names the readback. The length a scope was written
out for is watched: a cook that answers another length plans the graph again.
- A `where` whose condition the plan knows takes one alternative. What only the other alternative stands on is not
written out, and a call nothing reads that stands on any of it is not written out either (the nodes a node asset
leaves unread are written after every alternative of the plan is decided), so nothing reachable only through the
untaken alternative is cooked while the plan is made. A node of the graph itself that nothing reads is always
cooked.
- A `for` scope's `carries` (`{ [id] = { init, result } }`) chain the copies: copy 1 starts a carried value at `init`
and copy n+1 at what copy n's `result` lowered to; a `loop_result` read after the loop is what the last copy left.
- A step becomes one solver node. Its state is one record of device Columns per play (`g:state(name,
{ column = { type, count } })` allocates them zeroed). The columns a state allocates are stated through plan
values: a literal record, or a record a fold fills or constructs read off a list or an input (per-source columns,
counts from `list.count`, names from `geometry.names`), whose counts the plan cooks and lists in `report.waits`.
A `member` or `item` read inside the body of a value the plan knows (a count or a list a fold inside the step is
written out for) is answered while the plan is made. The step record carries `dt`, `time`,
`graphTime`, `index` and `first`. The body runs `substeps` times in each fixed step of `dt` seconds.
Pausing holds the state, a seek back replays from the start on a timeline with an end, and a changed
value the step reads begins it again.
### A plan is made once
A graph is planned once for the values its plan stands on, and every later cook of it reads that plan.
- A plan stands on the registry (`registry.revision()`), on the graph its plain inputs build, on the values of the
inputs it read (as keys), on the animated values it read at the clock, on the answers it waited on and the nodes
it cooked for a count or a list, and on the modules it required (by what their files hold). A cook reads a kept plan
after asking each of those again (`expand.stillStands`); a plan that no longer holds is made again, and a plan that
does is read.
- The body of a node asset is planned once for each structure its calls give it and kept (`expand.bodyCounts`): a
plan made again reads the kept body, and the nodes it writes out are the ones it wrote before. A kept body stands
while its asset's source is as it was and every definition it was written out from (the asset's own, and each op
and node asset it calls, at any depth) is the one the registry holds. A node registered again lets go of the bodies
that reach it and no other, so an edit of a node cooks again what the node reaches and nothing beside it.
- A Field that states what it computes (`proc.fieldc`'s `_expr`) is keyed by that statement and the operands it
names, not by the closure it samples with: a field written out again keys as it did.
- Each answer and node the plan waited on is held by the key of its cone, without a cook. The leaves a cone stands on
are named once, when the plan is kept (`expand.keyWatches`): the graph inputs it reads, the animated values, the
constants that hold a device value (a Geometry, a Column) or an asset reference, and the nodes that read live state;
the ops, their versions, the plain constants and the wiring are fixed by the plan and the registry. A read asks the
evaluator for the key of each leaf as it stands now (`opts._leafKeys`: a device value is keyed by the run it stands
on and the writes the run has had, an asset reference by what the asset's files hold) and compares
them with the keys the watch stood at. A watch whose leaves are unmoved holds and cooks nothing; a watch whose leaf
moved is cooked again, only its cone, and holds when it answers what it answered, and the keys it stands at are
renewed. A plan that cooked a node reading the clock, a play's input or a live run, or whose cone stands on a
solver or an op that always cooks, is not keyed: each of its watches is cooked again on every read.
- A plan made against its base, which it took the tables of, takes the base's leaf list, leaf index and the cone of each
node (`expand.keyWatches(plan, dynamic, base, reach)`; `expand.setWatchesCarried(false)` walks every cone). The cones of
the nodes the edit reached (`reach`, from `eval.moveKeys`) and of the nodes it let go of are dropped, and only the
watches whose roots hold no cone are walked; the rest keep the places they stood on. The places keep their numbers: a
leaf that no cone and no watch holds any more is a tombstone (`kind = "dead"`, key 0) that no list names, and when the
tombstones are at least 64 and outnumber the leaves the next plan walks whole. A registry that moved since the base
was keyed, a base that is not keyed and a plan that wrote its own tables are walked whole. `expand.counts()` answers
the plans carried and walked whole (with the reasons), the watches kept and walked, the nodes walked, the tombstones
and the compactions.
- A kept plan also holds the answer of each target it was cooked for (up to four a plan, and sixteen plans at a time),
filed by the keys of the target's cone: a cook of `proc.eval` that shares no node cache with its caller, finds the
cone unmoved and passes no `stats` or `cookRecord` is answered with a copy of the held answer (a Geometry or a Column
is cloned, a list or a record of them rebuilt), and cooks nothing. An answer holding anything else (a Handle, a
Field) is not held. The held answers are let go of with the plan.
- A plan is kept unless it registered a step solver or wrote a run of the device into itself (a `for` over
geometries). A plan that stands on a run of the device holds by the key of its cone, and is made again when the key
moves. `eval.planCounts()` answers how many plans were made and read (`made`, `reused`), the milliseconds each took
(`madeMs`, `reusedMs`), `notKept`, the plans that could not be kept by what stood in the way, `served`, the cooks
answered from a held answer, and `notKeyed`, the kept plans whose watches are cooked again on every read, by why.
- `engine` is the engine namespace, a constant. A member read or an item read out of constants, out of a record or a
list a call wrote, is the ref written there and costs no node; `engine.require` of a module named by a constant is
the module, required while the plan is made and asked for again each time the plan is read.
- The plan is made in slices of `budgetMs` (5 ms) when the caller can yield; a cook over the plan hands a frame back
after `cookBudgetMs` (100 ms).
- A cook that shares no node cache with its caller keeps none: nothing is keyed or filed, what the plan cooked while it
was made (or asked again) and what the cook makes stand in one table by node, and all of it is released as the cook
ends but what the answer reaches. A Field that is the answer keeps the runs its leaves read; they stand as long as
it does. What the answer reaches takes in the values a geometry's columns carry (a packed piece's `Mesh` record and
the geometry its `mesh` holds): a value names what it carries and does not own through `__proc_carried`. A value a
node of the cook made that the answer reaches only through what it carries is handed to the value carrying it
(`substrate.holdCarried`) before the answer is copied or let go of: it stands while any holder of it stands (a copy
of a holder is a holder) and goes with the last of them. A column that owns its values (a detail value) owns the
geometries and columns in it and lets go of them with itself, so what it carries is what they carry. A node that
lists what its passes made (`eval.ownedBy`, a fold or a `for` over geometries) lists the carried values its answer
reaches that the values it was handed do not, and none a column owns. A store keeps what its entries carry from
the release of a cook (it is among what the store holds), and a store shared by plays counts, among the values an
entry lets go of with the last entry holding them, only the values the entry owns: a carried value is its maker's.
- A node of a plan that stands is held to the contexts its producers operate in once, takes the label of its refusals
from one string, and takes its params prepared once when every one is a constant.
### A cook after a change is run from a list (`cookrun`)
A kept plan whose cone stands on numbers (graph inputs holding a number or a Bool) is cooked node by node while those
numbers move. The recording wraps the plan's cone re-check and the cook together: the first such cook after the plan
was made is recorded as a compute list and the next one is recorded beside it; the words that differ between the two
recordings (a dispatch param, a word of a write) must each be the value of one leaf in both, and become the slots of
the second list. A cook after that runs the list with the leaves' values in its slots, asks the recorded plan-time
read-backs again, runs the watches that are not nodes (modules, animated values, the registry stamp) and answers a
device copy of the list's columns (`cookrun.counts()`: `recorded`, `diffed`, `replayed`, `verified`, `mismatched`,
`readbackChecks`, `dispatches`, `refused` by why): no node is cooked and no buffer is allocated but the copy. What
makes a recording stand, and what ends it:
- Two recordings are one cook when they make the same dispatches over the same programs and workgroups, bind buffers
one for one, and write the same number of words; a run placed at another offset (`zero_base_*`, a write offset) is
a placement. A first cook that uploads a mirror of a CPU column makes the pair differ once, so the comparison is
tried again on the next cook (twice at most). A moved leaf that is no word of the list is refused by name, since its
value reaches the device by a write the recording did not hear.
- A cook that read the clock, a play's input or a live run, and an answer that cannot be copied are not recorded.
- A `cpu.call` is first handed the geometries it is given with their columns on the device standing in as columns that
answer their type, width, length and run (`substrate.structureOf`): a call that asks for no numbers (the names of the
columns a class holds, the types a join decides) moves nothing off the device and records no read-back. A call that
asks a stand-in for its numbers, fails, answers a stand-in, declares a return that is no plain value, or is handed a
geometry counted on the device or a part is made again over the drained inputs, and its read-back is recorded.
- Every node whose cook read back from the device (every `transfer.readback`, whatever the traffic of its thread
counted, and a drained `cpu.call`) is recorded as a read-back row: its inputs, whether they were drained, and a
digest of its plain outputs. A node that holds nodes of its own is asked again by their rows. After the list has run
each row is asked again; one that answers another value than it did ends the recording and the cook is made node by
node, answering correctly. A read-back whose outputs are not plain data is not recorded. A graph that reads a figure
that moves with its inputs (the bounds of a mesh, a volume) answers another value at every cook; one that reads the
whole number or the verdict that figure comes to (a power of two, 0 for a solid) answers the same.
- A count the device holds is not a read-back row: a node states the room it is allocated (`device.bound`) and the list
runs the count on the device inside that room. A count the cook states no room for is allocated twice what the cook
measured, in powers of two, and a room is never smaller than an earlier recording of the entry allocated
(`Run:room`). The room a count of a number takes is a decision of the cook: the list is run again for every number
that takes the same room, and for a number that takes another the recording is let go of and made again.
- A recording holds the guards its nodes state (`Run:guard`, `device.guard`, `device.bound` without a room): words of the
device the list writes, which are 0 while the cook's conditions hold. After a replay the words are asked for without
waiting (`substrate.deferredReads`, counted apart from `readbacks`), and the cook of the entry that follows reads
the answers that have landed (at most eight replays ahead; past that the latest words are read at once). A word other
than 0 lets the recording go, so that cook is made node by node and recorded again: it raises the refusal the
condition states, or allocates a count that outgrew its room four times the room (`counts().guarded`).
- The first replay of a list is held to the cook of the same values made node by node, column for column; a
difference ends the recording (`mismatched`). A list that dispatched a kernel declaring `determinism: float_order`
(a float sum whose threads arrive in any order) is held to f32 columns that agree to the rounding of a sum taken in
another order.
- A buffer a cook lets go of while it is recorded stays until the recording goes (`substrate.deferReleases`): the list
names it, and a replay writes it.
- A changed key of any other leaf (a Geometry input, a constant device value, an animated value), a plan made again
and a registry change let go of the recordings; at most six plans hold recordings.
- The recordings of two threads do not meet: a dispatch and a write belong to the recording of the thread that made it.
- The first replay answers a copy word for word onto runs of its own (`cookrun.deviceCopy`) and is held to the cook.
Every replay after it answers a share of the runs the list wrote (`cookrun.handOut`; `counts().shared`): a Geometry
clone, the columns of which share the runs, costs the caller nothing to make. The recording names every holder of
those runs (a clone of a share is named too); before the list runs again it moves the holders still out onto a copy
of the runs, so an answer a caller holds across cooks keeps the values it was handed (`counts().detached`). The held
answer of a plan is always a copy.
`cookrun.setEnabled(false)` cooks every cook node by node.
A `g:fold` of three passes or more over device state is planned as one `construct.fold_run` node (`foldrun`): the body
is built once as a flat graph with the inputs `state` and `index`, the first pass is cooked while a compute list
records it, and every later pass queues that list over a copy of the state, so a pass after the first cooks no node.
`foldrun.counts()` answers how many passes were recorded, replayed and cooked, and how often a body was refused a
list, by why. A body cooks each pass on its own when it reads `index`, when the state holds something the list cannot
follow (a record, a text, a readback), when a pass makes a state of another shape, or when the fold sits inside
another recording. `foldrun.setEnabled(false)` writes the folds of the plans made after out once per pass. The shape of
a state names each column it holds; a column carried as values (a record, a text) and a detail value stand by the
evaluator's key of what they hold (`eval.keyOf`), under which a geometry or a column a record holds stands as the runs
it reads and the writes they have had.
### How a step runs
A step solver plans its body once (`proc.eval.stepplan`), when the state takes its first step.
- A node of the body that depends on neither the state, the `step` record, an animated value nor the scene is STATIC:
it is cooked once, with the first step, and read as it stands by every later step (`opts.fixed` of the evaluator).
The plan reads only the nodes the next state stands on.
- What is left is the device work of a step. The first step of each distinct set of `step` values the body reads (its
`dt`, `first`, `progress`, ..., and the animated values it reads, as the device sees them: the f32 they round to) is
cooked with every dispatch and every write it makes into a device buffer kept as one compute list over buffers
the solver holds, and the next state written back into the state's own columns by the last dispatches of the list.
Every later step with those values queues the list: the dispatches read the state's columns where they stand
and write the next state into them. A step that runs from a list cooks no node and allocates no buffer.
- The columns of the state are written in place, so what a step presents is a copy of them made on the device.
- Up to six sets of `step` values are kept. A body the plan cannot stand on is cooked a step at a time with its
static nodes still read as they stand, and says why in `report.steps[<step id>].reason`: a node of the body reads
the clock, a play's input, a live run or the scene; the body reads values back from the device to the CPU; the
next state is not a record of Columns the state's own shape; or a `step` value the body reads takes a new value in
more than six steps (a body that adds `step.time` to a column, say).
- `play:report().steps[<step id>]` holds `steps`, `replays`, `recorded`, `interpreted` (steps cooked node by node),
`plannedDispatches` (what one list queues), `dispatches` (queued in all), `allocations` and `uploads`
(buffers made and uploads done by steps), `ops` (nodes cooked in steps, by op), `staticNodes`, `variants`, `mode`
(`replay` or `cook`), `reason` and `keyValues` (the `step` values of the last step).
- `stepplan.setPlanning(false)` makes the solvers begun after cook every step node by node, which is how a planned
body is held to the numbers a cooked one gives.
A state column the next state names as a Field, a number or a vector is evaluated into a column of the state's own
length and type.
### Constructs
A construct is a node the evaluator itself knows: `construct.member`, `construct.item`,
`construct.member_set`, `construct.column`, `construct.column_set`, `construct.column_remove` and
`construct.engine` and `construct.select` are cooked; `construct.pick`, `construct.compare` and `construct.where` are
written out in the plan (a compare of two constants is their answer, a `compare` otherwise calls the `math.lt`, `le`,
`gt`, `ge`, `eq` or `ne` node of its operator, and a `where` over a condition the plan does not know is a
`construct.select` node). `construct.select` chooses
`a` where `cond` holds and `b` where it does not: numbers, vectors and Fields element by element (`math.where`), and a
Geometry, a record or a text by a condition that is one value, which a condition over the elements of a column is not
(it is refused naming `g.pick`). A Column or a Geometry a construct reads from a record or a list is a clone of it.
`construct.expr` is the arithmetic of a graph as one node. The plan writes each operator (`+ - * /`, `math.*`, `vector.*`,
`g.compare`) as a `cpu.call` of a function of `fieldc.operators` and the `construct.item` that reads its result, and
after the flat graph is written out it gathers every connected set of them, with the `x`, `y`, `z` and `w` reads of their
results and the `g.where` choices between them that a condition per element makes, into one `construct.expr` whose `tape`
lists the steps (`{ fn, args }`, `{ fn = "where", choice = true, args }` or `{ lane, args }`, each operand a `leaf`, an
earlier `step` or a `const`) and whose `leaves` are the values the steps read. A condition is per element when it
stands on an element field or on a column of the elements of a class; the cook makes both alternatives of such a choice
and chooses between them element by element, so the plan decides nothing for it. A `construct.select` whose condition
is one value the cook answers, or whose alternatives are whole values, stays a node. Each call step runs through
`callrun.invoke`, the call `cpu.call` makes, and the node answers the last step. `flat.folds[id]` lists the flat id of the
call or choice each step stands for, and `describeFailure` reads a failing step at that call. A node the plan, a watch or
an output names, and a member written for another alternative of a choice than its reader, stays a `cpu.call` and a
`construct.item`; a member a decision names is the root of an expression of its own. An expression holds members a cook derives a number through (a replay of a recorded list works those out
again for other values of the graph's inputs) or members it does not, never both. `docs/specs/procgen-system.md` section
13.1 has the rules.
`construct.member` reads `count` and `keys` of a plain list or record (the number of members, the sorted names of a
record or the places of a list; a value that is nothing has none), `count`, `type` and `lanes` of a Field (the length
of the longest column its chain reads, its element type, its components; a Field that reads no column has no length,
and the read refuses naming `geometry.point.count`, `column.count` and `g.field.index({ domain })`), and `count`, `type` and `lanes` of a Column.
`construct.member` also reads `x`, `y`, `z` and `w` of a vector (a Field of that component for a Field, a Column of that
component for a Column of Vec2, Vec3, Vec4 or Quat), the classes of a Geometry, its `material` and its `names` (the
attribute names each class holds). On a class of a Geometry a read of a `detail` column answers the value the detail
holds. `count` of a Column or of a class of a Geometry is the live count when the elements are an appended or compacted
output (a Column of one word, which a kernel with a `domain_count` binding reads the live count through, and a kernel
without one runs over the capacity); `capacity` of either is the number of elements the run is allocated for. A record a
graph holds under a name a class goes by (`point`, `vertex`, ...) is read and written as a record: `x.point.name` is the
member of the member.
`construct.column`, `construct.column_set` and `construct.column_remove` take `edge` beside `point`, `vertex`,
`primitive` and `detail`: an edge is a unique point pair of the primitives, so the class counts them and a column
written on it is one value per edge. `count_set` takes the four stored classes.
`construct.column_set` writes a Column, a Field evaluated over the class it is written into, a vector or a number, a
record or a text (one value for every element, a constant Record or String column) or a list of records or texts (one
to an element, held in a Record or String column on the CPU); a record whose fields are numbers is a record, and a list
of numbers is a vector. A column that is the structure of a geometry (`point.id`, `vertex.pt`, `primitive.vstart`,
`vcount`, `curve` and `closed`) holds whole numbers: a number, a Field or a Column written to it is stored as an Int,
and so is a value written to a column that already holds Int. A detail value is held as itself, a record among them,
and `construct.column_remove` removes it. The
inputs of every construct take nothing as a value: an optional input a call left out reaches them as nothing.
`construct.scene { what }` answers where the graph stands in the scene it is cooked in (`entity`, `placement`,
`inverse`, `rotation`); it is read as `g.clock.placement` and the other names, and as `g.transform`.
### What the plan answers
A construct that answers while the plan is made cooks the nodes it reads (a wait in `report.waits`, a readback
named as one) and records the answer, which is what the play compares to decide whether the plan still stands.
- `check { cond } message`: a condition that is false refuses the play with the message.
- `has { geometry, class, name }` is the Bool of whether the class holds the column; `names { geometry, class }` is
the list of attribute names the class holds. A `for` over the names has one copy per name.
- `fold` with `stop`: after each pass but the last the stop is answered, and the fold ends at the first pass whose
stop holds. `"changed"` holds when the pass left its state as it found it (numbers by value, a Column or a
Geometry by what its runs hold: a device run read for the comparison is a readback wait). A stop function's scope
reads `fold_before` and `fold_after` and its result is the Bool. The report's fold scope states `copies`,
`stop` (`"changed"` or `"function"`) and `stopped`; each answer is a wait named "the stop of fold '<id>' after
pass <n>".
- A node asset called with plain inputs (an input with options, or of type Bool or String) is the graph its source
builds for those values (`builder.forInputs`); a play built for other plain inputs plans again.
Constructs cooked as nodes:
- `count_set { geometry, class, count }`: the geometry with the class's element count set. A number sets it; a
live count Column (an appended output's) becomes the live count over the capacity it carries; arithmetic over a
live count (`kept.indices_count * 6`, a Field over live counts) is evaluated on the device as one word, and the
class is as large as that arithmetic reaches when every count it reads is full (`substrate.atCapacity`).
- `material_set { geometry, material }`: the geometry carrying the material (a Material value, an asset reference
or a name; nothing for none).
- `step_between { step, index }`: how far from 0 to 1 the step has come through the span of graph time the present
covers; `step_sample { step, value }` is the value, which a step body samples at the step's time.
The step record carries `dt`, `time`, `graphTime`, `index`, `first`, `start` (the time the step began at) and
`progress`.
An `element` ref is the Field the graph's `g.element` stands for: `id` and `P` are fields the field compiler fuses
into the chain they sit in; a column of the evaluation domain is `fieldc.element(field, class)`.
### Cooking
Every play cooks into one store of the world, filed by content key (the op, its version, the seed, the
params and the keys of the inputs): a node of any graph that cooks the same op over the same inputs
is cooked once. A value that holds an asset reference is keyed by what the asset's files hold, so
editing a module a node calls cooks the node again. The nodes a play pulls from and every step are the
play's own.
An output node answers a Handle (a table with a `release` function). `cpu.call` owns its list of
results: a Handle in it is released when the cook is superseded and when the play ends.
A present evaluates the pulls whose keys can have moved and leaves the others standing with the Handles they hold. A
pull stands when the plans made since it was evaluated reached none of the nodes its key stands on (`moveKeys` files
them), it was evaluated at this clock, with these inputs, at this registry revision and before the play last moved on
the device, no module or asset a cook read was edited, and its cone holds no solver, no graph used as a node, no live
read and no CPU registration. A pull whose cone reads the scene stands while the digest of the scene state the plan
reads (`play:liveDigest()`) is the one it was evaluated at. A play that takes over the node cache of the play it replaces
takes the Handles with it.
The device shape of a plan (`eval.timeOnDevice` with `keptWith`: which nodes keep the play's time from reaching it through
live values alone) is the sum of the share each node of its flat graph holds. A plan made against its base states it
from the base's: the nodes the plan wrote anew are taken again and every other node keeps its share
(`eval.deviceShapePatches()` counts these, `eval.deviceShapeWalks()` the walks of every node). The causes are listed in
node id order.
### What procgen keeps (`proc.memory`)
Every play, whoever started it, and each module-level cache that holds procgen state (the kept bodies, the world store
the plays share) is a keeper. `proc.memory.report()` answers `{ total, luau, plays, keepers, byPart }`: the bytes each
keeper holds, a play by part (`graph`, `plan.flat`, `plan.origins`, `plan.reuse`, `plan`, `nodeCache.values`, `nodeCache.entries`,
`nodeCache.solvers`, `play`) with the tier of the state, and the bytes the Luau heap holds (`luau`). A part states its bytes from
counts kept where its state is filed and let go of: a plan's flat node count, a node cache's entries and solvers
(`eval.cacheCounts(cache)`, kept by `fileEntry`, `dropEntry` and the store's writes), the kept bodies' flat nodes
(`expand.bodyCounts().kept`). The bytes one unit costs are `proc.memory.PER`, measured by walking plays of the 1x, 4x
and 16x graphs of the edit-speed corpus. The tier of a part is what letting it go costs: 0 nothing, 1 a table made again when read,
2 a node cache value or body cooked or planned again where an edit reaches it, 3 the play's edit state (the next plan is made
without its base), 4 the cache as a whole, 5 (`proc.memory.STANDS`) what a present stands on, which no budget lets go of.
▲ 0↑ born
# fieldc
A Field is a per-element expression. `fieldc` holds what a Field is made of and how it runs: a chain of templates over a position leaf, an element number, column reads and numbers is written as one WGSL function and runs as ONE dispatch of `kernels/map`, wherever it is consumed. Nothing dispatches per operator.
## Where a Field is evaluated
```luau
local column = fieldc.evaluate(field, domain, opts)
```
The one entry point a consumer of a Field calls (a kernel's input, an attribute write, an output): `device.kernel` for a Field bound to a column input, `attribute.write` for a Field value, and every other node that needs a Column.
- `field`: a Field, a Column, a `BufferRef` (a device column handed through `cpu.call`), a number or a vector record.
- `domain`: a number of elements, or a Geometry (its `opts.class` elements, points by default), whose attributes the leaves reading the evaluation domain bind.
- `opts`: `{ class, count, type, clock, where, label }`. `type = "Int"` writes i32 words (a Field tagged Int does so by default). `clock` is `{ time, frame, fps, duration }`: animated values are first sampled there. `where` names the caller in a refusal.
- Answers a Column of the domain's count. Raises, naming what refused, when the chain cannot be written as a function.
A chain holding an animated value or a live run answers a live column. `fieldc.liveRuns` finds it and `fieldc.tick(runs, clock)` rewrites it each frame: the animated values are sampled at the clock and written into the numbers the dispatch reads, and the dispatch runs again into the same run. An animated input therefore changes the result every frame without a cook.
## Fields the nodes make
| Construct | Module | Answers |
|---|---|---|
| `fieldc.field(v)` | `fieldc` | any socket value as a Field: Field, Column, BufferRef, number, vector, animated value |
| `operators.apply(name, ...)` | `fieldc.operators` | `math.*`: the plain result over numbers and vectors, a Field over a run or a Field |
| `fieldc.element(name, class?, width?)` | `fieldc` | what `g.element` stands for: `"id"`, `"P"`, or a column of the domain (`g.element.point.N` is `element("N", "point")`) |
| `leaves.elementId()` | `fieldc.leaves` | the element number, an Int Field (`g.element.id`) |
| `fieldc.position` | `fieldc` | the position leaf `p` (`g.element.P`) |
| `leaves.context(class, name, width?)` | `fieldc.leaves` | an attribute of the evaluation domain |
| `leaves.rand(seed, scope, id)` | `fieldc.leaves` | the unit draw, equal to `hash.unitFloat` to the bit |
| `leaves.call(fn, args)` | `fieldc.leaves` | a function of a prelude shader module of `kernels/` |
| `leaves.gather(column, index)` | `fieldc.leaves` | a run read at computed elements |
| `leaves.substitute(field, position)` | `fieldc.leaves` | the chain read at another point (`fieldc.withPosition`) |
| `leaves.index(domain)` | `fieldc.leaves` | an Int column of 0 to n - 1 |
A Field made by these answers `f.x`, `f.y`, `f.z`, `f.w` as the Field of that component. Components are Fields, so the Field of a vector is built from them with `operators.vec2`, `operators.vec3` or `operators.vec4`; Luau's `vector.create` takes numbers.
## How the leaves reach the kernel
A chain's leaves are columns. `kernels/map` states every leaf's shape in its `cols` table and reads up to as many leaves as one dispatch has storage buffers for (`compute_abi.directLimit`: 8 per stage on every target, less the table, the numbers and the output, so 5 for `map`) straight from the buffer of the leaf's own run, at no cost beyond the binding. A chain of up to 5 leaf columns is the one dispatch of `map` and nothing else. Leaves past those are copied into the table, one copy dispatch each.
`fieldc.bindSpec(spec, { direct })` makes the table; `direct` is the list of leaf slots read from their own runs (`compute_abi.directPlan(columns, limit)`), and the variant is made with the same list: `program:variant({ specialize = ..., direct = { cols = direct } })`. Without `direct` every leaf is copied into the table.
`fieldc.operators` folds numbers: every operand a number or vector gives a number or vector. A narrower operand widens to the widest (a number applies to every component, a shorter vector reads 0 where it has none). Division and remainder by 0, the square root and the logarithm of a value at or below 0 answer 0.
## The walk step contract (for `kernels/walk`)
A walk kernel takes its per-walker function as a Field spec, the way `kernels/map` takes `TEMPLATE`.
```luau
local spec = fieldc.walkSpec(step, { width = 2, count = walkers, done = stop })
-- a state of 8 floats: lanes 0 to 3 in `step`, lanes 4 to 7 in `high`
local wide = fieldc.walkSpec(low, { width = 8, count = walkers, high = high })
local bound = fieldc.bindSpec(spec)
-- specialize = spec.specialize; bind cols = bound.cols:buffer(), u = bound.u
```
- `width`: floats per state, 1 to 8. `step` is the next state's lanes 0 to 3, a Field of `min(width, 4)` components; above 4, `opts.high` is the Field of the `width - 4` components of lanes 4 to 7.
- The state is read through `fieldc.walkState(count, first?)`: `count` (1 to 4) lanes from lane `first` (default 0), all inside lanes 0 to 3 or all inside lanes 4 to 7. A state of 8 floats is read as two leaves, `walkState(4, 0)` and `walkState(4, 4)`. The nodes `field.walk_state({ count, first })` and `field.walk_index()` are these leaves in a graph.
- `step` and `high` are written over the state leaves, `fieldc.walkIndex()` (the step number k, from 0), `leaves.elementId()` (the walker's number), Column reads (one value per walker, or a gather at a computed index) and numbers.
- `opts.done`: an optional Field<Float> over the same leaves; the walker stops before step k when it is not 0.
- `spec.specialize` is a table of the specialization values the walk kernel takes, ready to hand to `device.kernel` / `program:variant` as its `specialize` value: `WALK_STEP` (`zero_walk_chain_step(s, s_hi, k, id)`), `WALK_STEP_HI` (`zero_walk_chain_step_hi(s, s_hi, k, id)`, or `s_hi` when `width` is 4 or less), `WALK_DONE` (`zero_walk_chain_done(s, s_hi, k, id)`, or `false` when there is no `done`) and `WALK_PRELUDE`, the module-scope WGSL that defines those functions, preceded by the preludes the chain calls. Lanes 0 to `width - 1` of `s` (lanes 0 to 3) and `s_hi` (lanes 4 to 7) are the state.
- The functions read its leaves through a `cols` columns binding (`cols_f32(slot, i, 0u)`, `cols_vec3(slot, i)`, `cols_i32` for Int runs) and its numbers through `u_get(k)`: the kernel declares those two bindings exactly as `kernels/map` does. `fieldc.bindSpec(spec)` answers the table (`cols`) and the numbers buffer (`u`) to bind; `release()` lets both go.
- A read past a run's end reads its last element.
## Other entry points
`fieldc.mapSpec(chain, count, domain, { position })` writes a chain as one function for `kernels/map` and `fieldc.runMap(spec)` runs it. `fieldc.compile` / `fieldc.column` write the same chain as a kernel of its own; they stay for the consumers that bind buffers directly.
▲ 0↑ born
# proc.fields
Value wrappers for the procedural-generation type system: `Field<T>`,
`PointCloud` and `Population`. Part of `@builtin::systems.procgen.proc`.
Typed values backed by plain strided Luau arrays.
- **`Field<T>`** — a lazy per-element value. `Field.constant(t, values)`,
`Field.fn(t, stride, f)`, `Field.buffer(t, stride, buf)`; `:sample(id)` →
the `stride` components, `:materialize(count)` → a flat `count*stride` array.
- **`PointCloud`** — the value a `.procNode` input declared `PointCloud`
receives (`proc.substrate.toPointCloud`); every other input, and every op,
takes points as a `Geometry` (`proc.geometry.new`). `PointCloud.new(positions)`
(flat stride-3 array); `:count()`, `:position(id)`, `:setAttribute(name, field)` /
`:getAttribute(name)` / `:attributeNames()`.
- **`Field.transform(offset, orient, scale)`** — a per-element transform whose
three parts are fields of their own, so each compiles to the device;
`:transformParts()` hands them back.
Every value carries a `__proc_type` tag that `proc.types.typeOf` reads
(`"Field<Vec3>"`, `"PointCloud"`, `"Population"`).
▲ 0↑ born
▲ 0↑ born
# proc.geometry
CPU geometry container (positions/indices/normals/uvs) + bounds, normal recompute, and renderer.mesh.create args.
Part of `@builtin::systems.procgen.proc`.
▲ 0↑ born
▲ 0↑ born
# proc.graph
The graph data model and its definition entry: the nodes a graph holds, the normalized `InputRef` edges between
them, its inputs and outputs, and its JSON (de)serialization. Required as
`@builtin::systems.procgen.proc.graph`; the top-level `proc` module re-exports its entries as `proc.define`,
`proc.graph(id)` (`new`), `proc.fromJSON` and `proc.op`.
`define(function(g) ... end)` builds a graph in the call form, whose authoring surface is `proc.builder`: nodes
called by asset identity (`g.points.disc({ ... })`), inputs typed by their defaults (`g:input`), sockets that take
Luau's operators, members, columns and `for`, and the declarations a graph states about itself: `g:describe` (what it
makes), `g:output` (what it gives), `g:family`, `g:timeline`, `g:still` and `g:cost`. The result is a graph with
`surface = "call"` that records each node by the guid of the asset it calls.
`refusedInput(graph, name, value)` names why a value is not one an input holds. `fromJSON(doc)` reads a graph
document back and `graph:toJSON()` writes it.
Part of `@builtin::systems.procgen.proc`.
▲ 0↑ born
# proc.hash
Deterministic, portable scoped hashing for the procgen evaluator.
`scoped(seed, graphId, compositePath, nodeId, channel, elementId) -> number`
folds its arguments with 32-bit FNV-1a (via an exact 16-bit-split `mul32`)
and a final avalanche, producing a bit-identical `u32` on native, WASM, and
Android. `unitFloat(...)` maps the same scope to a value in `[0, 1)` for
deterministic random draws.
No `math.random` / `os.time` / float transcendentals — determinism is the
contract. The per-element scope (`elementId`) keeps a given element's draw
stable as unrelated element counts grow.
▲ 0↑ born
▲ 0↑ born
# proc.noise
Deterministic, portable value/gradient noise. Part of
`@builtin::systems.procgen.proc`.
Built on the integer-lattice hash in `proc.hash` (`latticeCell2/3` /
`unitCell2/3` — the corner hashes of one cell, each bit-identical to the
matching `hashInts` call): every lattice corner's random value/gradient is
integer-derived, and the only
continuous math is the quintic fade + linear interpolation. No float
transcendentals and no `math.random`, so the result is bit-identical on native,
WASM, and Android — which is what makes a procedural texture reproducible
everywhere. Exposes `value2`/`value3`, `gradient2`/`gradient3`, and `fbm`.
▲ 0↑ born
# proc.ops
Op-family loader: registers every procedural op against the registry.
Part of `@builtin::systems.procgen.proc`.
▲ 0↑ born
# proc.parity
The machinery behind the procgen rewrite's parity harness (`procgenParity`), its sheet tests (`procgenSheet`) and the `procgen.coverage` tool.
- `parity.json`: `entries` holds one entry per current op. Each names its new composition (`new`, node ids as `nodebox.name`, or `nodebox.*` for any node of a nodebox), its class (`exact`, `meaning` with an `invariant`, `gone` for a named decision with a `note`, `unmapped` where the sheet names no counterpart), and `tolerance` the comparison reads. An entry that states `decided` (the text of the ruling that accepts the difference between the op and its node) keeps running: the value differences its comparison finds are reported as accepted, under the verdict `decided`, and are held to nothing, while a node that raises and the cost findings stay findings (`accepting`, `decidedLines`). `budgets` states, per op, the read-backs its node may make (`{ "<op>": { "readbacks": n } }`); an op without one may make as many as the current op does. `types` maps a socket's type to the fixture a case builds for it. `cases` holds a case per op that needs more than the defaults and names only content the library ships. A parity suite states the cases whose inputs are its test fixtures in the `cases.json` of its folder (`{ "cases": { "<op>": case } }`), since the library ships its test content nowhere; `cases()` is both together, one case per op, and an engine whose library leaves the suite out finds no case for those ops.
- A case states `params`, `fixtures` (a fixture by input socket), `node` (the new node a multi-node entry compares against), `rename` (`inputs`, `params`, `outputs`: the new node's names where they differ, `false` leaves a name out), `compare` (`bounds`: a Geometry is held to its bounds alone, for an output whose element layout differs by design; `ramp`: two ramps are held to the values they read at the same positions), `sink` (`asset` or `handles`: the op writes to the world, each side under its own name and what it wrote is compared and removed), `world` (entities spawned for the run; `scene` takes the run's root scene, `settle` waits until the op runs on them), `tolerance` (bounds that replace the global ones for this case, `{ attribute = 0.05 }`: for an op whose values come from the device's sampler, whose fixed-point weights and decode answer within a fraction of a percent of the exact blend a node computes), `ignore` (outputs left out), `unreachable` (the sentence that says why the harness has no fixture for the case; the entry then counts as a harness error) and, for an op that is a solver over a timeline, `suite` and `describe` (the group of an existing suite that holds its comparison; `deferred` leaves that group to the suite's own run and states the entry pending, for a group that costs the suite's whole time a second time inside this one).
- A fixture is a name (`mesh`, `points`, `curve`, `positionField`, or a type named in `types`), a current op run on its own fixtures (`{ op, params, inputs, output }`; the new side builds it through the node that replaces the op when the entry for that op names exactly one, so a ramp, a group, a prefab, a raster or an Sdf a node reads is one that library made; `shared: true` makes both sides take what the current op makes, for inputs the entry under test is held to and the op that made them is not), a node run on its own fixtures (`{ node, params, inputs, output }`, a new node by id), a list (`{ list }`), a record (`{ record }`) or a literal (`{ value }`). A case's `newInputs` are fixtures only the new side takes, for an input the current op states as a param (a texture the node reads where the op names the asset).
- `sheet.json`: the composition sheet's expectations: the four ops, the 23 kernel families with their tallies (a kernel asset `<kernel>_<pass>` belongs to the family `<kernel>`), the layers, the L2 ids as the core package names them, `universalUses` (the L2 nodes every row may call), one row per node family (the kernel families, L2 nodes and ops a node of the row may name; `authorKernel` for rows whose nodes run a kernel the author names; which nodes may stand on one kernel family alone), and the reported numbers.
`scan()` finds the new parts in the library: the nodes of every built-in nodebox wherever its package keeps it, the kernel families with their pass assets, and the modules that sit beside nodes. `analyze(src, ...)` reads a node's source: the nodes it calls (`g.<name>(...)`), the ops it names (`op.<name>`), the kernel families it names, a structural hash per call, and WGSL text in its strings. `rowViolations(scan, node)`, `dispatchViolations(scan, node)` and `directCallers(scan)` state what a node breaks of the sheet. `valuesHold(result)` reads a comparison's values alone, which is what lets an existing suite run on the nodes of an op whose cost is over. `assess(scan, entry)` states where an entry stands (`decided`, `unmapped`, `pending`, `ready`); `compare(scan, entry)` runs a ready entry's current op beside its node; `checkCase(entry)` runs the current op on the entry's case. `redirect(map, fn)` runs `fn` with `registry.get` answering a new node for each op id in `map`; `runSuite(name, redirect)` runs an existing suite's cases through the test framework's own runner, and `runSuiteOnNodes(scan, name, ops)` runs it as it stands and with its ops answered by their new nodes. `report(opts)` is the coverage report.
A comparison states cost as separate numbers (`result.cost`): `oldMs` (the current op's CPU time), `planMs` (the evaluation outside the nodes' own cooks: expanding the call form, lowering refs and constructs, scheduling, with the plan's share of the waiting taken out), `cookMs` (the CPU time the nodes' cooks spend, with their share of the waiting taken out), `waitMs` (the time in `task.wait`, a read-back's frames, shared between the cooks and the plan by the read-backs each made), `dispatches` (device dispatches queued), `readbacks` (read-backs the substrate counted) and `frames` (frames that passed during the evaluation). A Field in an output is read over the Geometry its side was given, a Column by its values, and a Bool answers 1 or 0. The cost ratio of `tolerance` is applied to `cookMs` against `oldMs`; the read-back count is asserted against the op's budget on its own, so a read-back removed from a node shows as `frames` falling to zero. Times are the median of three runs; counts are the last run's.
A comparison states its values and its cost apart. `result.valueFindings` are the differences in what the node answers and `result.costFindings` the cost and read-back findings (`result.findings` holds both). `verdict(assessed, result)` reads them as `value` (`equal`, `differ`, `missing` for an attribute, layer or column the new output lacks, `raised` for a node that failed on the case's inputs, `harness` for a case whose inputs could not be made, `skipped`, `decided`) and `cost` (`within`, `over`, `none`); `summary(scan)` counts every entry under exactly one of `pass`, `costOver`, `differ`, `missing`, `raised`, `harness`, `skipped` and `decided` and reports the same per family (the nodebox of the entry's first new node) with the median and worst cost ratio.
What is compared where the two libraries carry a value differently: a Geometry by its live counts (a domain allocated at a capacity compares its live front), a Raster layer by layer at the grid's cell centres, a Texture by size, channel count and texels (bytes scaled to 0 to 1 and held to half a byte step), a count the new library holds on the device as a one-word column by its value, a Population against the instanced points by instance count, and a list of numbers against a list of records of numbers as one layout finding that says whether the numbers agree.
▲ 0↑ born
# proc.placement
Where a graph stands in the world, as a 4x4 matrix held as 16 numbers column
after column: the layout `entity:worldMatrix()` and
`camera.viewData().viewProj` answer in.
A graph is evaluated in its own space, whose origin is the entity it stands on.
A Generator draws what it makes under its entity, so the graph's origin stands
where the entity does, and an effect built at the origin works wherever it is
played. A node that pulls from the scene (`scene.fetch`, `terrain.ground`,
`sim.scene_colliders`) reads that origin from the scene the evaluation hands it,
`ctx:scene().origin()`, and takes what it reads from the world into the graph's
space. A graph evaluated standing on no entity has the world's origin for its
own (`proc.placement.IDENTITY`).
`mul`, `inverse`, `point`, `scale`, `rotation`, `turn` and `conjugate` are the
matrix and quaternion steps such a node takes.
▲ 0↑ born
▲ 0↑ born
▲ 0↑ born
▲ 0↑ born
# proc.records
Record kinds: the values a graph moves that become assets when saved (a material, a mesh, a texture, a user's own).
`proc.record({ name, asset, applies?, fields?, live?, release?, save?, load? })` defines a kind; a `.procRecord` asset's
`init.luau` returns one. A record carries its kind's guid as `kind`; the node that writes it states the kind on the
geometry's detail attribute `records` (`{ [guid] = reference }`), and output, export and import find the definition
there. `save` and `load` default to the asset type's own `graph.write` and `graph.read`. A kind whose `live` value holds
something (a scratch asset) states `release`, which output calls once nothing stands on the value.
Part of `@builtin::systems.procgen.proc`.
▲ 0↑ born
# proc.registry
The operation registry: register/lookup/schema/search/help for procedural ops.
Part of `@builtin::systems.procgen.proc`.
▲ 0↑ born
# proc_field_noise (compute shader)
2D value-noise field: `dst[i] = value2(seed, coords[i] * freq)`. One GPU thread
per element. Re-implements `proc.noise.value2` — the integer FNV-1a lattice hash
(`proc.hash.hashInts`) is bit-exact-portable, so the `u32` hash here matches the
Luau `bit32` hash exactly; the quintic-faded bilinear interpolation runs in `f32`
(vs the CPU's `f64`), bounded to `f32` rounding.
▲ 0↑ born
# proc_field_scale (compute shader)
Element-wise scale of an `f32` field: `dst[i] = src[i] * scale`. One GPU thread
per element. Reads a storage buffer (`src`), writes a storage buffer (`dst`), and
multiplies by the `scale` params uniform.
▲ 0↑ born
# proc_raster_apply_accumulator
Folds one tile's fixed-point accumulator into a layer, clearing the
accumulator for the next batch.
`proc.raster`'s `Raster:resolveAccumulator(layerName, scratchName, work?)`
runs this after `proc_raster_edge_add` has folded in what a neighbour pushed
across the seam, then refreshes `layerName`'s apron so the next step reads a
consistent neighbourhood.
▲ 0↑ born
# proc_raster_apron
Refreshes one tile's apron ring of one layer from the neighbour tile that
owns those cells, one dispatch per (destination tile, neighbour direction).
`proc.raster`'s `Raster:exchangeAprons(name, work?)` drives it, over every
neighbour pair a work set touches. Any iterative raster kernel op whose
per-cell result depends on a previous step's neighbourhood needs its written
layer's apron refreshed this way between steps, or a later step reads a
neighbour's stale interior instead of what that neighbour actually computed.
▲ 0↑ born
# proc_raster_copy
Copies one raster tile into another, cell for cell, apron included.
`proc.raster`'s channel set uses it for checkpoints, clears, and layer copies.
A checkpoint is a copy of every live channel tile taken at a stack index, so
an edit below that index restores the copy instead of re-running the ops
above it. A clear is the same copy with a zeroed tile as the source. A layer
copy duplicates a raster's cells so a mutating op leaves the original untouched.
`dst`'s length decides how many cells are copied, so the two tiles must be the
same shape.
▲ 0↑ born
# proc_raster_edge_add
Adds a neighbour tile's apron-ring fixed-point accumulator into this tile's
interior, one dispatch per (destination tile, neighbour direction).
`proc.raster`'s `Raster:resolveAccumulator(layerName, scratchName, work?)`
drives it, before folding the accumulator into `layerName` and refreshing
`layerName`'s apron. Droplet and thermal erosion both accumulate this way —
any tiled kernel op whose per-cell result can spill across a tile seam needs
the same fold.
▲ 0↑ born
▲ 0↑ born
# proc.substrate
The substrate a graph carries: a geometry, its four domains (point, vertex, primitive, detail) and the named columns over them. A point is a position and an identity; a vertex is one primitive's use of one point; a primitive is a run of vertices (a face, a curve or a volume); detail is the geometry itself, one element. An attribute is a named, typed column over one domain. A numeric column is a `substrate.createBuffer` handle, the engine's one buffer primitive, so the handle a column holds is the one a kernel runs over and a compute shader binds; a column of a type a buffer cannot express (text, a reference, a record) is carried as values and placed on the device interned. `Geometry`, `Column` and `Domain` are the value types; `fromArrays`, `fromPositions`, `fromCurves` and `fromPointCloud` build a geometry. `man proc.substrate` lists every function with its arguments.
Part of `@builtin::systems.procgen.proc`; reached as `proc.substrate`.
▲ 0↑ born
▲ 0↑ born
# proc.types
The procgen type system: core type tags + socket assignability for graph validation.
Part of `@builtin::systems.procgen.proc`.
▲ 0↑ born
▲ 0↑ born
# yaml.module
YAML decode + encode for authored engine content. Pure Luau, no engine
dependencies — a general parsing library, not data-system plumbing.
```luau
local Yaml = require("@builtin::modules.yaml")
local doc = Yaml.decode(vfs.read(path))
vfs.write(path, Yaml.encode(doc))
```
## Supported subset
- Block mappings (`key: value`) and block sequences (`- item`), nested
to any depth.
- Compact sequence-of-mapping entries (`- path: x` / ` kind: y`).
- Flow collections — `{ a: 1, b: 2 }` and `[1, 2, 3]` — single line.
- Plain scalars with type inference: `~` / `null` → nil, `true` /
`false` → boolean, numeric literals → number, everything else →
string.
- Single- and double-quoted strings (`''` escapes a literal quote
inside a single-quoted string; `\n` `\t` `\r` `\\` `\"` inside a
double-quoted one). A quote opens a quoted scalar where a scalar
begins — after `key:`, after a `- ` entry marker, or inside a flow
collection — so a plain scalar carries an apostrophe as data
(`note: the world's rules`). Quoted keys are read in flow mappings.
- `#` comments — honored outside quotes, so a `#` inside a quoted
string is kept as data.
- Literal (`|`) and folded (`>`) block scalars, with a chomping
indicator (`-` strips the final newline, `+` keeps every trailing
blank line, none keeps one newline) and an indentation indicator
(`1`-`9`: the content's indent past the parent's), in either order
(`|-`, `>+`, `|2`, `>1-`). A block scalar's content is literal text:
quotes (balanced or not), `#`, `key: value` and `---` inside it are
data.
- An optional leading `---` document marker.
## Loud-error contract
Constructs outside the subset above raise instead of misparsing:
anchors/aliases (`&`, `*`), tags (`!`), directives (`%`),
multi-document streams (a second `---`), a block-scalar header
it cannot read (`|x`, `|-+`, `|0`), tab indentation, and inconsistent indentation (both block structure and
block-scalar bodies). Every error carries the offending line number
(`yaml: line N: ...`), so a bad config points straight at the line to
fix.
Tab indentation is rejected everywhere — including block-scalar body
lines whose leading whitespace contains a tab. A tab after the first
non-space character of a line is content and passes through.
## API
- `Yaml.decode(text: string): any` — decode a YAML document into a
Luau value (table, scalar, or nil for an empty document). Raises on
malformed input or an unsupported construct.
- `Yaml.decodeWithComments(text: string): (any, boolean)` — decode as
`decode` does, and report whether the text carries a comment by the
decoder's own rule: a `#` at a line's start or after whitespace, outside
a quoted scalar and outside a block scalar. Re-encoding the value drops
such comments, so an editor that writes a document back reads this first.
- `Yaml.encode(value: table): string` — encode a Luau table as YAML
(block style, two-space indent, alphabetically sorted keys). Raises
on values YAML can't represent (functions, userdata, non-string
mapping keys).
A top-level empty table encodes to an empty document, which decodes
back to `nil` (the YAML empty-document ambiguity).
▲ 0↑ born
# assetType.shared.settings
Resolves four questions about an asset's typed settings, all answered from its
type's `type.yaml` — nothing here is found by convention.
```lua
local Settings = require("@builtin::assetTypes.assetType.shared.settings")
-- or, from inside the assetType runtime:
local Settings = require("@builtin::assetTypes.assetType.shared").settings
```
## Where does a type's schema live?
`type.yaml` names it in a `settings:` block:
```yaml
settings:
values: "settings.yaml" # where an instance's own values live
schema: { type: "settings.luau" } # one schema for every asset of this type
```
or, for a type whose settings differ per instance (an importer):
```yaml
settings:
values: "settings.yaml"
schema: { instance: "settings.luau" }
```
or, for a type whose schema is a table its instance's own source declares and
whose values live elsewhere (a component: its `public` fields, valued on each
entity it is on):
```yaml
settings:
schema: { instance: "init.luau", export: "public" }
```
Whether an instance may lack its schema file is what the type's `files:`
block says of that file. Listed under `files.optional`, it may be absent, and
an instance without it declares no settings — the importer type lists it
there, since most importers take none. Listed under `files.required`, or not
listed at all, it must load, and an instance missing it is a broken
declaration and is logged.
`schema.type` is required at the folder relative to the TYPE's own
`.assetType/` directory (one schema shared by every instance).
`schema.instance` is required relative to each INSTANCE's own folder (`init`
resolves the instance's own identity; anything else is a sibling module under
it). Without `schema.export`, the file is required as a module and its return
value IS the schema. With `schema.export`, the file is run in an environment
of its own and `export` is read from a returned table carrying it, or else
from the globals the file assigned — a component's `public` is such a global.
That environment reads through to `_G` and takes every write itself; its own
`_G` is itself, its `declare` and `computed` record into it, and its `task`
records spawned, deferred and delayed work instead of starting it. Every other
global is the real one, so a call the file makes through one runs as it does
anywhere. The schema is the declared table's Field descriptors. A schema that
resolves to nothing says why through `Settings.schemaProblemOf(ref)`: the file
could not be read, compiled or run, or declared no such table. An instance
schema is read again when its file, its type's `type.yaml`, or the entry file
of any module the asset `require`s changes. `Settings.blockFor(typeName)`
returns the raw block; `Settings.schemaForType(typeName)` and
`Settings.schemaOfInstance(ref)` resolve the two schema shapes respectively;
`Settings.schemaFor(ref)` tries the type-level schema first, falling back to
the instance-level one. `ref` naming an assetType asset itself — e.g.
`asset.resolve("texture", "assetType")`, rather than an instance of it —
reports that type's own declared schema, the same one every instance of it
reports.
## Where do an asset's values live?
`settings.values` names a file living beside the asset's own content —
`Settings.valuesPathOf(ref)` joins it onto the asset's path.
`Settings.valuesOf(ref)` / `Settings.writeValues(ref, values)` read and write
it; this module only names where it is. A type that declares its schema with
no `settings.values` keeps no values in the asset — a component's live on each
entity it is on — so `valuesOf` answers nil for it, `asset.create` refuses a
`settings` table for it, and `asset.setSettings` refuses a write.
`Settings.recordedWrite(path, text)` writes a file the way every values file
is written, and an editor uses it for any other file it writes. It writes the
whole contents, records the write into the open undo operation when one is
open (so undoing it restores what the file held), and raises when the write
does not land.
## In what order are a schema's fields declared?
Every `Field.*` constructor stamps its descriptor with `declaredOrder`, one
more than the last. Luau evaluates a table constructor's entries in the order
they are written, so `Settings.orderedNames(schema)` lists a schema's fields —
and a struct's subfields — in the order the source writes them.
A values file, or a legacy `.metadata` settings block, may hold an older
spelling — a renamed field, an older enum member. Every read passes what is
stored through the type's `normalizeSettings` (`Settings.normalizeFor`), so
it answers in the fields the schema declares today, and writes nothing. The
next local edit of that asset rewrites the file in the declared spelling
(`Settings.healLegacyIfNeeded`).
`asset.setSettings(ref, patch)` writes a partial patch:
`Settings.resolvePatch(ref, patch)` normalises it, merges it over the current
values and applies the type's derived defaults for the fields it changed, and
`Settings.writeValues(ref, values, patch)` applies the corrections, checks the
schema and writes the file.
## How is a new asset's settings resolved at creation?
`asset.create` calls `Settings.resolveForCreate(typeName, name, destPath,
creationOpts, callerSettings, importerSettings)` before running the type's
`onCreate`, checks the result against the schema, writes it to the values
file, and hands `onCreate` a COPY of it as `opts.settings` (`Settings.
cloneValues`) — never the same table the values file gets, so nothing
`onCreate` does to its own copy can reach what was stamped. `onCreate` reads
settings from `opts.settings` and never writes to it. `asset.create` also
seeds the change baseline with the stamped values
(`Settings.seedChangeBaseline`), so the values file's own first dispatch
reports nothing changed and the payload `onCreate` encoded is the only
encode a create runs.
The layers, lowest priority first, each overriding the ones before it field
by field:
1. The schema's own declared defaults (`Settings.defaultsOf`).
2. The type's own `settingsDefaults(name, opts, asked)` hook, when its
`behavior.luau` exports one. `asked` holds what layers 3 to 5 name, so a
default can follow a field one of them sets.
3. The Preset Manager's matching row (`Settings.presetLayerFor`).
4. The creating importer's settings for this type.
5. The `settings` table the caller passed to `asset.create`.
6. The type's own `settingsDerivedDefaults(values, explicit, changed)` hook,
when its `behavior.luau` exports one — a field whose default follows
another field, set only when no layer from 2 to 5 named it.
7. The type's own `settingsCorrections(values, name, opts)` hook, when its
`behavior.luau` exports one — applied last, once every other layer has
merged.
Layers 2 to 5 are each read through the type's `normalizeSettings`, so a
layer naming a field in an older spelling sets the field it names today.
`asset.create`'s top-level settings alias accepts those older names too
(`Settings.acceptedFieldNames`).
Layer 7 is `Settings.applyCorrections(typeName, values, explicit, context)`,
the one corrections path every settings write takes — this create chain, and
`Settings.writeValues` for every later write (`asset.setSettings`, a type's
first-encode seed, the legacy heal). `explicit` names the fields the writer
asked for: the caller's `settings` table here, the patch for
`asset.setSettings`. A correction lands SILENTLY on a field the writer did not
name — nobody asked for that value, so overriding it just states a fact. A
correction that contradicts a field the writer named REFUSES the write,
naming the field, the value asked for, and why it cannot hold — the same as
every other place this codebase refuses an explicit request it cannot satisfy
(a `Field.checkSchema` violation, a BC7 alignment error) rather than silently
rewriting it into something else. So `asset.setSettings(tex, { compression =
"rgba16" })` on a mipped texture stores `generateMipmaps = false` beside it,
while `asset.setSettings(tex, { compression = "rgba16", generateMipmaps = true })`
is refused.
`Field.checkSchema` runs once, over the result AFTER the corrections hook has
applied — a bad value nothing corrects refuses the creation.
### Four hooks, four different jobs
Every hook is an optional export on a type's `behavior.luau`, alongside
`onCreate` / `onChange`. **None is ever handed a table to mutate** — each
reads what it's given and RETURNS a table of what it wants changed (or nil);
`resolveForCreate`, `resolvePatch`, `applyDerivedDefaults` and
`applyCorrections` are the only places that write to the values, by merging
what a hook returns the same way every other layer merges.
- **`normalizeSettings(values) -> values`** maps every older spelling a
stored file or a write may carry to the field and member the schema
declares now, returning a new table; a table naming both an older field
and its replacement keeps the replacement. The companion export
**`normalizedSettingNames = { ... }`** lists the older field names it
accepts beyond the schema, so a write naming one is not refused as an
undeclared field. `texture.assetType`'s maps the fields its earlier schema
stored to `colorSpace` + `compression`, and the older `filter` keywords to
`linear` / `nearest`.
- **`settingsDerivedDefaults(values, explicit, changed) -> partial?`**
answers the values of fields whose default follows another field. At
creation `changed` holds every field; on a write it holds the fields the
write moved. A returned field applies only where `explicit` — the fields
the writer named — does not name it.
- **`settingsDefaults(name, opts, asked) -> partial?`** supplies a DEFAULT
the schema itself has no syntax for — something only knowable at creation
time, like a naming convention, a fact read off a payload the caller handed
in, or a value that follows a field a higher layer sets (`asked` holds what
the Preset Manager, the importer and the caller name, the higher one's
value where two name a field). It runs early (layer 2 above), so any later
layer — Preset Manager, importer, caller — freely overrides it.
`texture.assetType`'s returns the `textureType` the name suggests (a
`_normal`-suffixed name is a `normalMap`), `compression = "uastc"` and
`maxDimension = 4096`, with `rgba8` for a `textureType` of `sprite` that a
higher layer names; for a pre-encoded ZTEX payload it returns the colour
space, compression and sampler the payload's own header carries. Either
way, an explicit `settings = { ... }` from the caller still wins.
- **`settingsCorrections(values, name, opts) -> (partial?, reasons?)`**
enforces a CROSS-FIELD CONSTRAINT the schema itself has no syntax for — a
fact about one field that depends on another field's resolved value, true
regardless of which layer produced that value. It runs last (layer 7
above), over the fully merged `values`, and again on every later write of
the values file. Nothing is left unmet in what gets
validated and stamped — but a value the caller explicitly asked for is
never silently rewritten to satisfy it; `applyCorrections` refuses instead
(see above). The optional second return, `{ [field]: reason }`, supplies
the "why" for that refusal message; a field with no reason falls back to a
generic one. `texture.assetType`'s returns `{ generateMipmaps = false,
maxDimension = 8192, colorSpace = "linear" }` whenever the RESOLVED
`compression` is high-precision (the encoder stores one verbatim, linear,
and refuses a mip chain or a downsample for it outright), paired with a
reason naming the compression, and `nil` otherwise.
## What does a declarer declare?
`Field.settingsOf(identity, mode)` embeds another declarer's schema as a
struct field without restating it. The identity names one of two kinds of
declarer, and `Settings.schemaOfDeclarer(identity)` tells them apart by what
`identity` resolves to:
- **An assetType** (e.g. `"@builtin::assetTypes.texture"`) — embeds the
schema every instance of that type carries, resolved via
`schemaForType`. This is the type-level shared schema from the section
above, not a per-instance one — an assetType has no settings of its own
to declare per instance.
- **Anything else** (an importer, or any other asset that declares settings
of its own) — embeds that asset's own schema, resolved via
`schemaOfInstance`.
Every schema this module serves has already run its `Field.settingsOf`
fields through this resolution — a caller reading `schema.someField.fields`
never sees an unresolved `declarer` with no `fields`. Resolution descends
into an already-literal struct's `fields` and a list's `element` the same
way, so a declarer embedded several fields deep is still reached. A declarer
that embeds itself, directly or through another declarer, stops one level
short of the loop instead of recursing forever.
## What are the defaults?
`Settings.defaultsOf(schema)` reads every field's declared `default` into a
plain table — a field declared without one is absent from the result rather
than holding a `nil` value.
▲ 0↑ born
# assetType.shared.typeYaml
A type's `type.yaml`, parsed. Every block a type declares in its `type.yaml`
(`settings:`, `inspector:`) is read through this module, so one file read and
one YAML decode serve all of them.
```lua
local TypeYaml = require("@builtin::assetTypes.assetType.shared.typeYaml")
local folder, identity = TypeYaml.folder("material")
-- "/zero/source/libs/@builtin/assetTypes/material.assetType", "@builtin::assetTypes.material"
local doc = TypeYaml.of("material")
print(doc.suffix) -- ".material"
print(doc.inspector.type) -- "inspector.luau"
```
- `TypeYaml.folder(typeName)` resolves the assetType and answers its folder
path and identity, or `(nil, nil)` when no such type resolves.
- `TypeYaml.of(typeName)` answers the decoded document, or nil when the type
does not resolve or its `type.yaml` does not decode to a table. The decode is
cached per type and read again whenever the file's content version moves, or
the type resolves to a different folder.
- `TypeYaml.described(typeName)` lists the files a type's assets must describe:
each file its `type.yaml` declares with a `described` content check, with the
keys and headings the check names and each key's byte limit. `TypeYaml.PLACEHOLDER`
is the marker a template puts where a description is yet to be written.
- `TypeYaml.describeBlock(typeName)` answers the `describe:` block of the
`type.yaml`: the type's own `guidance` on what a description covers and the
identities of `examples` whose descriptions show the shape.
▲ 0↑ born
# asset_change_dispatch (module)
Installs `_G.__zero_dispatch_asset_change`, the Luau half of the
asset-type **change-callback** system. The engine calls it once per VFS
source write (via `zero_scripting::ffi_callbacks::fire_asset_change_dispatch`,
queued as `VfsMutation::AssetSourceWritten`).
## What it does
1. Receives the written VFS path.
2. Walks up the path to the enclosing `<name>.<type>/` typed-asset folder
(deepest registered-type suffix wins, so the direct owner of the write
is chosen for nested typed assets).
3. Loads that type's `<type>.assetType/behavior.luau` via
`@builtin::assetTypes.assetType.shared.ref.loadTypeModule`.
4. If the module exports an `onChange` function, calls
`onChange(ref, change)` where `ref` is the typed `AssetRef` for the
changed asset and `change = { path, asset, type }`.
The engine runs each dispatch as its own yieldable task, so a hook can wait
on what it reads: a `vfs.read` of bytes only the blob store holds (a binary
file on the web build) fetches them. A hook that reads only what is resident
finishes inside the dispatch that started it. The writes routed to one asset
react in the order they were routed: a write routed while that asset's hook
is waiting queues behind it and reacts once the hook has finished.
This is the type-level analogue of the component-centric
`onAssetReload(field)` fan-out: components react to assets they
*reference*; an asset *type* reacts to writes inside its own *instances*.
## The source-write seam
A write inside a typed asset is also offered to the importer type's shared
layer (`onAssetWrite(assetPath, path, origin)`), the way a write with no
enclosing asset is offered to `onLooseWrite`. The asset's `import` record names
the file it was imported from (by guid and path); a write to that file re-runs
the importers on it, in place; the write that arrived is the only input. A type whose own `onChange` already keeps its retained source current
(a texture re-encoding its image, a sound clip its audio) sets
`reencodesRetainedSource = true` on its module, and the seam leaves that edit
to it.
## Re-entrancy
While an asset's `onChange` runs, a dispatch for that same asset made from
the hook's own coroutine is suppressed, so a handler that routes a write
back into its own folder inline doesn't recurse. A write the hook makes is
routed on a later frame like any other and reaches `onChange` again once
the hook has finished. `onChange` handlers must therefore be **convergent**: diff the meaningful
state before acting and short-circuit when there's nothing to do (the
`dynamicAsset` example only regenerates when the prompt actually changed
and bails while a generation is in flight). Idempotency is the contract,
exactly as it is for services' `start`/`stop`.
## Related
- `@builtin::assetTypes.assetType.shared.ref` — owns `loadTypeModule` + the per-type `ref` method
dispatch.
- `assetTypes/assetType.assetType/template/behavior.luau` — the documented
template that shows how to author `ref`, `global`, and `onChange`.
- `assetTypes/dynamicAsset.assetType` — the reference type that uses
`onChange` for prompt-driven regeneration with version control.
▲ 0↑ born
# asset_create
`asset.create` — the single, generic "instance a new asset of an existing type"
API. The same entry point scripts and agents use; there is deliberately no tool
wrapper, because agents author in Luau and call `asset.create(...)` directly.
```lua
local inst = asset.create("material", "my_metal", { base_color = { 0.8, 0.7, 0.2 } })
-- → AssetRef: inst.path == "/zero/source/my_metal.material", inst.guid, plus the
-- material type's ref methods.
```
`asset.create` makes a new *instance of an already-registered type* by running
that type's `onCreate(name, opts)` behaviour hook (declared in its
`behavior.luau`) and writing the produced files to
`/zero/source/<name>.<typeName>/` — the one authored location, in every mode.
While play runs, the play-mode write lock takes that write onto the **play
shadow**: live in the session, listed by
`vfs.playShadowPaths()`, and promoted or discarded on a guarded play-exit. The
asset created in play therefore carries the same identity, path and `require`
spelling it has in edit, and the session decides whether it stays.
A create whose output the world **reproduces** on every load — one made from a
component or a scene entrypoint — writes to the copy-on-write runtime store at
`/zero/runtime/assets/<identity>.<typeName>/` instead, so the code that rebuilds
it each load is its only source and the saved manifest never carries a second
copy.
## Placement
Four `opts` keys are consumed by the framework before the type's `onCreate`
hook runs, and steer where the instance lands:
| Key | Effect |
|-----|--------|
| `folder` | A relative subfolder under the source root: `/zero/source/<folder>/<name>.<typeName>/`. Groups a generator's output instead of accumulating it at the source root. |
| `into` | A resolved container ref (`.toolbox` / `.package`) to author INSIDE — lands at `<container>/<name>.<typeName>/` and registers as a member. Edit-mode only. |
| `dest` | An absolute destination path, owned by the caller (an importer building a `<name>.bundle/`). |
| `overwrite` | Re-author an existing destination in place, keeping its `.meta` guid so every reference stays valid. |
The `name` is always the asset's bare identity — the path goes in `folder`:
```lua
local rock = asset.create("mesh", "rock", { positions = p, indices = i, folder = "terrain/props" })
-- → rock.path == "/zero/source/terrain/props/rock.mesh"
```
`dest` and `into` both take precedence over `folder`. `folder` decides the
asset's identity, so it applies in every context: the runtime store is flat and
spells that identity in one folder name.
```lua
-- the same call from a component's awake()
-- → rock.path == "/zero/runtime/assets/terrain.props.rock.mesh"
-- → rock.identity == "terrain.props.rock", as it is from an execute
```
Types without an `onCreate` hook fall back to cloning their verbatim
`template/` skeleton to the same destination, so `create` works for every type.
Returns the created asset's `AssetRef` (`.path` / `.guid` plus the type's ref
methods — the same interned instance `asset.resolve` returns) on success and
raises (via `error`) on bad arguments or a failing `onCreate`.
Installed onto the FFI `asset` namespace by the prelude (`M.installInto(asset)`).
See `docs/specs/runtime-asset-copying.md`.
▲ 0↑ born
# declaration (module)
A type's `type.yaml`, decoded and held: the one reader behind every question
about a type as a whole — its files, whether it broadcasts its subassets, how
its instances are pictured.
## Exports
- `M.of(typeName) -> table?` — the decoded declaration, or nil when no type
answers to the name, it has no `type.yaml`, or the file does not decode (the
last with a warning naming the type). Shared: read it, do not write to it.
- `M.forget()` — drop every held declaration. The change dispatcher calls it
for every write inside an `.assetType` folder and for every removed type.
▲ 0↑ born
# asset_events
Per-asset runtime for declared asset events. An assetType's `behavior.luau`
declares an `events` schema the same way a component declares one; this
module turns that schema into the live objects an asset's ref fires and
listens on.
## Three faces, one Signal — keyed by the asset
The construction is `component_events`': one `Signal` per declared event,
reachable through a private fire-capable table, an owner-side emitter the
type's own behavior fires with, and a subscribe-only facade every other
holder of the ref sees. The facade exposes `connect` / `once` / `wait` and
has no `fire` at any key, so firing authority stays with the type.
What differs is the owner. A component event belongs to one instance on one
entity; an asset event belongs to the **asset**, so the runtime is keyed by
the asset's stable guid. Every resolver of that guid subscribes to the same
Signals, and a ref that is reclaimed and re-resolved re-attaches to the
subscriptions already there — the same reason `asset_ref`'s `runtime` table
is guid-keyed rather than stored on the envelope.
Both tables reject an undeclared event name: the private table's metatable
raises on a bad key and the facade raises on a bad key, so a typo surfaces
at the subscribe call instead of returning nil.
## Declaring events on a type
```lua
-- <name>.assetType/behavior.luau
local M = {}
M.events = {
changed = { payload = { value = Field.number(0, NoSync) } },
}
M.ref = {
poke = function(self)
local emitter = require("@builtin::assetTypes.assetType.shared.events").emitter(self.guid)
if emitter ~= nil then emitter.changed:fire({ value = 1 }) end
end,
}
return M
```
```lua
-- any holder of the ref
local ref = asset.resolve("@builtin::…")
ref.events.changed:connect(function(p) print(p.value) end)
```
## Exports
- `M.forAsset(guid, eventSchema)` — the runtime for one guid, built on first
use and reused after. A later call with a different schema reconciles:
surviving events keep their Signal and their subscribers, new events get a
fresh one, removed events are disconnected and dropped.
- `M.facade(guid)` — the subscribe-only view, or nil before a runtime exists.
- `M.emitter(guid)` — the owner-side `:fire` view, or nil before a runtime
exists.
- `M.has(guid)` — whether a guid holds a live runtime.
- `M.teardown(guid)` — drop one asset's runtime, disconnecting subscribers.
- `M.teardownAll()` — drop every runtime. Called on an engine mode flip so a
play-mode subscription does not survive into edit.
- `M.liveGuids()` — every guid holding a runtime, sorted.
## Notes
- Subscriptions made through the facade are ordinary `signal.module`
connections and disconnect the same way any other does.
- The runtimes table is held strongly, and its lifetime is bounded by the
mode flip that clears it.
▲ 0↑ born
# assetType.shared.inspector
Which inspector a type declares for its assets, and the declaration that file
returns. The editor's Inspector (`edui.inspector`) reads it to add, replace and
order the sections an asset shows.
## The block
A type names its inspectors in `type.yaml`, with either key or both:
```yaml
# the Inspector for every asset of the type, a file in the type's own folder
inspector: { type: "inspector.luau" }
# a file each asset may ship in its own folder, for what that asset declares
inspector: { instance: "inspector.luau" }
# both
inspector: { type: "inspector.luau", instance: "inspector.luau" }
```
`type:` is what selecting an asset of the type shows. `instance:` applies to
the things an asset declares, never to the asset's own view: a component
names `instance: "inspector.luau"`, and a component that ships that file
shapes its card on every entity it is on. Each value is a `.luau` file name.
A block that is not a map, names another key, names neither key or names a
file that is not `.luau` is malformed; the assetType validator reports it as
`assetType.inspector_block`, and a `type:` file missing from the type folder
as `assetType.inspector_file_missing`.
## The declaration
The file returns a table:
```luau
return {
-- sections this inspector adds (before Operations)
sections = {
{ id = "summary", title = "Summary", build = function(ctx)
return { ctx.widgets.text{ text = "mode is " .. tostring(ctx.settings.mode) } }
end },
},
-- a generic section id -> the section shown in its place
replace = { tags = { id = "tags", title = "Labels", build = function(ctx) return {} end } },
-- section ids, first to last; ids it does not name follow in default order
order = { "summary", "identity" },
}
```
A section is `{ id, title, icon?, open?, build(ctx) }`; `build` returns the
section's body as a list of edui nodes. Any member of the declaration other
than `sections`, `replace` and `order` must be a function (a helper the file
exports). `material.assetType/inspector.luau` is a full example.
The Inspector rebuilds an asset's sections when the asset's files change. A
section that shows state held outside those files names it with a
`signature(ref)` member: text that moves when that state moves, read every
time the Inspector checks for a rebuild, so it has to be cheap.
`renderFeature.assetType/inspector.luau` folds in whether the feature runs
and its live settings, so a feature enabled or reconfigured from a script
shows in the Inspector as it happens.
## API
```luau
local InspectorDecl = require("@builtin::assetTypes.assetType.shared.inspector")
InspectorDecl.blockFor("material") -- { type = "inspector.luau" }
InspectorDecl.blockProblem({ other = "x.luau" }) -- why a block is malformed, or nil
InspectorDecl.declaresTypeInspector("material") -- true: the asset view reads one
local decl, problem, file = InspectorDecl.declarationFor(matRef)
local cdecl = InspectorDecl.instanceDeclarationFor(compRef)
InspectorDecl.shapeProblem(decl) -- nil when well-formed
```
`declarationFor(ref)` (and `typeDeclaration(typeName)`) read the `type:` file:
- `(nil, nil, nil)` when the type names no `type:` inspector;
- `(nil, problem, file?)` when the block is malformed, or the file is missing,
raises while loading, or returns the wrong shape — the problem names the
file and what is wrong with it;
- `(declaration, nil, file)` otherwise.
`instanceDeclarationFor(declarerRef)` reads the `instance:` file in the
declaring asset's own folder, and answers `(nil, nil, file)` when that asset
ships none.
▲ 0↑ born
# asset_on_register
Drives the optional `onRegister(self)` assetType lifecycle hook — the surface that lets an asset instance run code exactly once when it first registers, operating on the per-instance ref (`self`). This is what makes "write an asset into the world → it takes effect live, no restart" work for content that registers into a runtime registry (editor panels, and anything else that opts in).
A type opts in by exporting `onRegister` from its `behavior.luau`. The hook fires exactly once per instance:
- On `engine.onWorldLoaded` — a batched sweep over every world instance of every type that implements `onRegister`, yielding every 32 instances so a large instance count never stalls a frame. The @builtin library (`/zero/source/libs/`) is excluded.
- On live `asset.create` — the create path calls `M.fireFor(ref)` on the freshly-created instance.
A per-VM once-guard keyed by the instance's VFS path ensures the two triggers can't double-register the same instance.
## Exports
- `M.fireFor(ref, onRegisterFn?) -> boolean` — fire one instance's `onRegister(self)` hook exactly once. Resolves the type's hook from `ref.type` when `onRegisterFn` isn't supplied. Idempotent: a second call on the same instance is a no-op. Returns true if the hook fired this call.
- `M.sweep()` — batched sweep: fire `onRegister` once for every world instance of every registered type that implements it. Yields, so it must run inside a task. Safe to call repeatedly.
- `M.install()` — install the world-loaded sweep. Idempotent. Runs the sweep on a system task so it ticks through pause and never blocks the caller.
▲ 0↑ born
# previewRecipe (module)
How a type makes its instances' pictures. A type declares it in its
`type.yaml`, under `preview:`, and every instance's `preview.png` is produced
by the recipe named there.
```yaml
preview:
kind: surface # instantiate | image | surface | custom
mesh: sphere # surface: the builtin mesh the asset is painted on
```
| `kind` | The picture is | Needs |
|---|---|---|
| `instantiate` | the asset stood in the studio through its own `instantiate` | — |
| `image` | a file the instance already holds, fitted to the frame | `source.file` or `source.files` |
| `surface` | the asset painted on a builtin mesh | `mesh` |
| `custom` | whatever the type's `behavior.luau` `ref.preview` returns | — |
`angle: [yaw, pitch]` (degrees) states the bearing the camera looks from, for
any kind that renders.
The `preview:` block says how the picture is PRODUCED. Whether search embeds
that picture is a separate declaration — an `image` chunk under
`indexing.content` naming `preview.png` — so a type states both, and a type
that states one without the other gets exactly what it stated.
## Exports
- `M.of(typeName) -> PreviewRecipe?` — the recipe `typeName` declares, or nil
when it declares none or no type answers to that name. Read once and held;
a block naming an unknown `kind`, or missing what its kind needs, is refused
with a warning naming the type.
- `M.forget()` — drop every held recipe. The change dispatcher calls it, with
`declaration.forget()`, for every write inside an `.assetType` folder and for
every removed type, so an edited `type.yaml` is read again.
▲ 0↑ born
# assetType.shared.ref
Builds the metatable every `AssetRef` (the `{ __ref, type, name, guid, identity, path }` envelope returned by `asset.resolve` / `asset.ref`) carries. Replaces the per-call C metatable that used to live in `crates/zero_scripting/src/ffi/bindings/asset.rs::attach_asset_ref_metatable`.
## Public API
```lua
local assetRef = require("@builtin::assetTypes.assetType.shared.ref")
-- Called by the Rust factory (push_asset_ref_handle) via
-- _G.__build_asset_ref_proxy after the envelope fields are set.
assetRef.build(envelope) -- attaches AssetRefMT, returns the table
```
The module exposes no invalidation surface — discovery is registry-driven and hot-reload is transparent (see "Discovery and hot-reload" below).
## Method dispatch order
`ref.foo` is resolved in this order:
1. **Default methods** — `getSource` / `getBytes` / `getText` / `exists` / `inspect` and the lazy `meta` property. Per-type definitions cannot override these (engine contract from gh#1889).
2. **Per-type `ref` table** — discovered through `asset.resolve(<typename>, "assetType")` + `require(<identity>.behavior)`. The lookup resolves to whichever `<typename>.assetType/` folder is currently registered for the typename (built-in, user, or library-vendored) and pulls its `ref` methods.
3. **Nil** — unknown key, same shape as the original C metatable.
## Adding methods for a new asset type
Drop a `behavior.luau` into the type folder:
```
src/lua/lib/assetTypes/<typename>.assetType/
├── type.yaml -- structural spec (existing)
├── behavior.luau -- ref + global behaviour (new)
└── template/ -- placeholder body (existing)
```
`behavior.luau` returns:
```lua
return {
ref = {
--!desc Per-instance method on every `<typename>` AssetRef.
myMethod = function(self, ...) ... end,
},
global = {
-- Reserved for the asset.<typename>.* surface — not consumed
-- yet; included so the contract from gh#1889 lands cleanly.
},
}
```
Within `ref` methods, `self` is the AssetRef envelope: `self.path`, `self.identity`, `self.guid`, `self.type`, `self.name` are available. Engine-required fields (`guid`, `path`, `identity`, `type`) must not be redefined.
## Discovery and hot-reload
Type discovery uses the asset registry, so it doesn't matter where a `<typename>.assetType/` folder lives — engine built-ins under `@builtin::assetTypes.<X>`, user types dropped under `/zero/source/<X>.assetType/`, or types vendored inside an imported library all resolve identically. The dispatcher walks `asset.resolve(<typename>, "assetType")` → canonical identity → `require <identity>.behavior` on every miss; the asset registry is the cache for the resolve half and Luau's `_LOADED` is the cache for the require half.
Hot-reload is transparent. When a `<typename>.assetType/behavior.luau` is rewritten through VFS, the engine's `invalidate_require_cache_for_state` re-runs the module and mutates the cached `_LOADED[<identity>.behavior]` table in place. Dispatch reads `mod.ref` on every access (rather than caching a sub-pointer), so existing AssetRefs see the new methods on the very next access — no Rust-side hook into this module is needed, and no invalidation API is exposed for callers to call.
The same registry-driven path also handles type **registration** correctness: dropping a new `<typename>.assetType/behavior.luau` updates the asset registry, the next `asset.resolve(<typename>, "assetType")` returns the new ref, and AssetRefs of that type immediately gain the new methods.
## Why Luau, not Rust
The asset surface needed per-category behaviour (`material:getProperties()`, `bundle:instantiate(entityId)`, `tool:run(args)`) and the old shape couldn't grow without an FFI change per method. Owning the proxy in Luau means:
- Type-specific methods ship as content (the `<typename>.assetType/` folder), not engine code.
- Adding a new asset type drops in a new folder; no Rust rebuild.
- Methods can call any other Luau global (`vfs`, `asset`, `Material`, `entity`, etc.) directly, instead of going through narrow FFI primitives.
## Consumers
- `crates/zero_scripting/src/ffi/bindings/asset.rs::attach_asset_ref_metatable` — the Rust factory that pushes envelopes invokes `_G.__build_asset_ref_proxy` once per push.
- Every Luau call site of `asset.resolve` / `asset.ref` / `entity(id).component.X.material` / any binding declared `AssetRef<...>` — they all receive envelopes built through this module.
▲ 0↑ born
# asset_ref_shapes
Publishes what every `AssetRef<category>` answers to, so a member read on an
asset-typed value is checked against the category's own surface.
An asset category's per-instance surface is authored: a
`<name>.assetType/behavior.luau` declares `M.ref = { ... }`, and each category
declares its own. The members an `AssetRef<inputMap>` carries are therefore
knowable only to `inputMap` itself — no fixed set of types covers the ones a
world defines, and the checker has no way to guess them.
This module reads each registered category's `M.ref` table from its source
(never executing it), renders it as a Luau table type, and hands the set to
the engine. From there a `ref:method(...)` on an asset-typed value resolves
against the category's real surface: a name it does not carry is reported
along with the ones it does.
## Results shaped by the asset
Some results are shaped by the asset rather than by its category —
`inputMapRef:activate()` answers one handle per binding THAT map declares, a
set that is authored and differs per map. A category states those by
declaring `refShapes`, and this module asks it once per instance and
publishes the answers keyed by asset identity. `types/assetType` documents
the authoring side.
## When it publishes
Publishing is driven by use: every entry point that produces diagnostics
calls `ensure`, which is a no-op once the set is current. A world load, a
write inside a type definition, and a write inside any asset whose container
computes shapes each mark it stale, so the cost lands on the next check and
only if one comes.
The sweep is complete and replaces what was published before, so a category
whose type is removed stops being published.
## Reading what the checker believes
From a call site, a type that is correct and one that was never published are
the same absence of a diagnostic. `published()` tells them apart:
```lua
local shapes = require("@builtin::assetTypes.assetType.shared.refShapes").published()
shapes.categories.inputMap
--> "{ activate: () -> any, controls: () -> { any }, ... }"
shapes.returns["@builtin::inputMaps.default"].activate
--> "{ crouch: Handle, interact: Handle, jump: Handle, ... }"
```
▲ 0↑ born
# texture
An **image asset** — the raw bytes of a `png` / `jpg` / `jpeg` / `hdr` /
`ktx2` / `webp` / `bmp` / `tga` file. A loose image written into the world is
imported into a `<stem>.texture/` container (see "Internal layout").
> New to how a texture exists as an asset, as CPU-side pixels, and as a live GPU
> texture (and how `renderer.texture.*` and render targets move between them)?
> Read `core/resource-model` first.
`texture` is a first-class assetType (not a bare plural-dir category) so
that:
- a texture has a stable asset **identity** and an explicit `asset_type`
link, and
- a material that references a texture records a real `.refs` edge to it
— which is what lets the CPU-residency system cascade from a material to
the textures it actually uses.
## CPU residency vs GPU residency
A texture can be resident two fundamentally different ways, and they must
not be conflated:
- **CPU residency** (this assetType): the raw image **bytes in memory** —
readable, hashable, decodable. You can operate on the bytes.
- **GPU residency**: those bytes uploaded to the device as a
GPU texture the renderer binds/samples. Byte-level reads are **not
possible** on the GPU variant — that is simply how the device works. GPU
residency is managed separately and is **not** part of this assetType.
Same texture identity, two separate residency states, two separate
operation sets, two separate reference counts.
## Internal layout
A `.texture/` holds these files, and its child set is open. `asset.create("texture",
name, ...)` and the **texture importer** (which creates through it when a loose
raster image lands in `/source/`) produce this shape:
```
<name>.texture/
source.<ext> the retained source raster — see below
source.ztex the full-resolution texels a cook dropped some of — see below
settings.yaml the texture's settings — see below
data.ztex canonical pre-encoded + pre-mipped payload (ZTEX magic)
.metadata what data.ztex holds and was encoded from
```
`asset.create` keeps a copy of the image bytes it was given as `source.<ext>`;
the importer moves the uploaded image in under its own name instead
(`foo.png` stays `foo.png`). Either way it is the texture's retained source:
rewriting it re-encodes `data.ztex` from the new pixels, and a rewrite carrying
the bytes `data.ztex` was already encoded from changes nothing.
`data.ztex` is the **preferred primary**: when present the renderer resolves
and uploads it verbatim — no re-mip. When absent the primary falls through to
a raster image in the folder, so a container holding only an image file reads
that image.
That fallthrough takes the lowest-named raster image in the folder and never
the rendered preview, which is made from the payload. A
container that lost its encoded payload and holds no other image has no
primary: `asset.diagnose(ref)` reports `noPrimaryFile`, `asset.validate(ref)`
reports `texture.no_payload`, and `renderer.texture.loadCpu` refuses it.
`asset.primaryFile(ref).path` names the file the resolution landed on; the
asset-system guide covers `diagnose`.
#### `settings.yaml` settings
A texture's settings are declared in `settings.luau` and live in its
`settings.yaml`. `asset.create` stamps them when the texture is made, and
`texRef:setSettings({ ... })` (or `asset.setSettings(ref, patch)`) changes
them. Editing an encode-relevant one re-runs the encode against the retained
source and rewrites `data.ztex`.
```lua
local tex = asset.create("texture", "hero_albedo", {
bytes = png, ext = "png",
settings = { maxDimension = 1024 },
})
tex:setSettings({ uastcRdo = "medium" })
```
| key | values | default | meaning |
|-----|--------|---------|---------|
| `textureType` | `color` / `normalMap` / `mask` / `sprite` | `color`, or guessed from the name | what the texture holds, recorded in the payload. `normalMap`: tangent-space normals — when `generateMipmaps` builds the chain, every level of it, the base included, is renormalised to unit length; with `generateMipmaps = false` the one stored level is the source as given, and the chain the GPU builds from it at upload is a plain box filter. `mask`: linear data (roughness, metallic, occlusion, height, masks). `sprite`: a 2D image with transparency. A new texture's type is guessed from its name — `_normal` / `_nrm` / `_n` / … → `normalMap`, `_rough` / `_metal` / `_ao` / `_height` / `_mask` / … → `mask` — unless it is given. |
| `colorSpace` | `srgb` / `linear` | follows `textureType`: `srgb` for `color` and `sprite`, `linear` for `normalMap` and `mask` | how the stored values are read when sampled. `srgb` for albedo/emissive/base-color, UI and sprites; `linear` for normal/RM/AO/height/mask data. |
| `compression` | `uastc` / `rgba8` / `rgba16` / `rgba32f` | a new texture: `uastc`, a `sprite` `rgba8` | what the texels are stored as — see "What each compression stores" below. |
| `effort` | `fastest` / `fast` / `thorough` | `fastest` | how long a `uastc` encode searches for each block — see "Block compression" below. |
| `uastcRdo` | `none` / `low` / `medium` / `high` | `none` | how far a `uastc` encode moves from the source to make a payload that compresses smaller on disk and over the network — see "Block compression" below. |
| `generateMipmaps` | `true` / `false` | `true` | bake the full mip chain into `data.ztex` at encode time. |
| `maxDimension` | integer, `1` to `8192` | a new texture: `4096` | the longest side the encoded texture may have; a larger source is downsampled to fit, and a `uastc` texture's longest side is the largest power of two within it. A value past 8192 is refused. |
| `filter` | `linear` / `nearest` | `linear` | sampler filtering recorded in the payload. A material slot and a UI image widget sample the texture with it. A post-process effect's texture property, a render pass's `@texture` input and a compute shader's `texture_2d` binding sample through a sampler of their own (the effect's `postprocess.setSampler`, the pass's, the compute shader's named `sampler` binding), which this setting does not change. |
| `wrapU` | `repeat` / `clamp` / `mirror` | follows `textureType`: `clamp` for `sprite`, else `repeat` | how a material samples the texture past its left and right edges: tiled, the edge texel held, or mirrored. Recorded in the payload. |
| `wrapV` | `repeat` / `clamp` / `mirror` | follows `textureType`: `clamp` for `sprite`, else `repeat` | the same, past its top and bottom edges. |
| `anisotropy` | `world` / `1` / `2` / `4` / `8` / `16` | `world` | samples taken along a surface seen at a grazing angle. `world` follows the `render.anisotropy` world setting (`renderer.setAnisotropy`); a number fixes it for this texture, clamped to what the device supports. A `nearest` texture samples with 1. A write may give the number itself (`anisotropy = 8`); it is stored as the string. Recorded in the payload. |
| `alpha` | `input` / `premultiplied` / `none` | `input` | what the encode does to alpha. `input`: kept as the source has it (straight alpha). `premultiplied`: colour multiplied by alpha — decoded to linear light, multiplied and re-encoded for an `srgb` texture, multiplied as stored for a `linear` one — the source a material with the `premultiplied` blend mode (`src + dst * (1 - src.a)`) composites. `none`: every texel opaque, colour kept. |
| `alphaIsTransparency` | `true` / `false` | follows `textureType`: `true` for `sprite`, else `false` | fill every fully transparent texel with the average colour of its non-transparent neighbours, spreading outward, so bilinear filtering and mip averaging never pull black into the edge of a transparent area. Alpha stays 0. Runs before `alpha` and before the mips. |
| `keepCpu` | `true` / `false` | `false` | keep the decoded pixels in the CPU store after the GPU upload — see below. |
The type's colour space, `alphaIsTransparency`, `wrapU` and `wrapV` are
stamped when the texture is created, and a later write that changes
`textureType` moves them with it, unless the same write names them.
A cook that drops texels — a `maxDimension` smaller than the source, or a
`uastc` cook resampling a side to a power of two — keeps the full-resolution
input as `source.ztex` (one RGBA8 level) when the container holds no source
image, so raising `maxDimension` later re-encodes from every texel. A cook
that drops nothing keeps nothing extra. A texture made from raw pixels whose
cook dropped nothing keeps its payload as its only source, so a settings
change re-encodes from that payload read back with straight alpha: what an
earlier encode discarded stays discarded, after `alpha = none` its alpha is
opaque, and a `uastc` payload re-encoded is the blocks' texels, not the
pixels first given.
A new texture's settings start from the texture type's own defaults —
`uastc` with a `maxDimension` of 4096; `rgba8` for a `sprite`; a 16-bit PNG
or `.hdr` source left to the schema's `rgba8` unless a layer names a
compression — and a world's Preset Manager rows
for `texture` override them (a world that ships at 2048, say), then an
importer's settings, then the `settings` the create names.
`filter`, `wrapU`, `wrapV` and `anisotropy` say how the texture is sampled,
not what its texels are. A write that changes only those rewrites the
payload's header (`renderer.texture.withSampler`): every level keeps its
bytes, and no preview is rendered, because no pixel changed. Every material
slot samples through its own texture's state, so two textures on one material
can wrap and filter differently. UI image widgets honour `filter` and draw each
image over its own bounds, so the wraps apply to materials. A post-process
effect, a render pass `@texture` input and a compute shader sample through
samplers of their own, so none of the four settings reaches them.
`keepCpu` is a lifecycle setting, not an encode input — it never re-runs the
encode. It acts on an already-materialised texture right away instead: `true`
loads the decoded pixels into the CPU store when the texture is GPU-resident,
`false` drops a resident CPU copy.
The `data.ztex` payload is a `ZTEX` magic, a 32-byte header and the mip
levels; the CPU encoder is exposed to Luau as
`renderer.texture.*` and consumed by the texture importer and this assetType.
#### What each compression stores
| `compression` | In `data.ztex` | On the GPU | Limits |
|---|---|---|---|
| `uastc` | UASTC 4x4 blocks, whole mip chain, one byte per texel, each level compressed for storage. | Transcoded at upload to a block format the adapter samples: `Bc7RgbaUnorm[Srgb]` where it samples BC textures (desktop native, desktop WebGPU), ASTC 4x4 where it samples ASTC and not BC (phones, a browser on one), ETC2 RGBA8 where it samples only ETC2, each one byte per texel; RGBA8 at four bytes per texel on an adapter that samples none of them, and for a texture with a side under four texels. | each side is resampled to the nearest power of two (halfway rounding up), nearest-neighbour for a `nearest` texture. |
| `rgba8` | From an image with `generateMipmaps = false`, a `maxDimension` the image fits within, `alpha = input` and `alphaIsTransparency = false`: the source PNG / JPEG / WebP itself, decoded at load. Otherwise raw 8-bit RGBA with the encoder's box-filtered mip chain. | `Rgba8Unorm` / `Rgba8UnormSrgb`, four bytes per texel. | none |
| `rgba16` | Raw 16-bit unsigned normalized, one level. An image is decoded at its own depth: a 16-bit PNG keeps every bit, an 8-bit channel `v` is stored as `v * 257`. The frames of an animated image are decoded at 8 bits. | `Rgba16Unorm` where the adapter grants 16-bit normalized textures (native); half float (`Rgba16Float`, about 11 significant bits) where it does not (every browser). | one level, no `maxDimension`, `linear` only. |
| `rgba32f` | Raw 32-bit float, one level. An image channel is stored as its value over the channel's maximum. | `Rgba32Float`. Where the GPU filters 32-bit float, every consumer samples it. Where it does not (every browser): a material slot draws the shader's declared default; a UI image draws nothing; a render pass naming it as an `@texture` input is skipped; a compute shader that declares no sampler binds it and reads it with `textureLoad`, and one that declares a sampler refuses the dispatch — each logging once, naming the texture and its format. CPU reads (`:load()`, `renderer.texture.decode`) return every bit on every device. | one level, no `maxDimension`, `linear` only. |
A high-precision compression turns `generateMipmaps` off, `maxDimension` to
8192 and the colour space to `linear` when the same write does not name them,
and refuses a write that names a contradicting value.
#### Block compression
`compression = "uastc"` stores every mip level as UASTC blocks — sixteen bytes
per 4x4 texels — a format no GPU samples directly but every one can be handed:
at upload each level is transcoded to a block format the adapter samples, BC7
on a desktop GPU and ASTC 4x4 on a phone's (ETC2 on one that samples ETC2
alone), so one payload is resident at one byte per texel on each, and RGBA8 on
an adapter that samples none; `renderer.textureMemory().blockFormat` names the
format this adapter holds it in. Only the levels a device keeps are
transcoded: under a texture-size cap (`renderer.setMaxTextureSize`) a texture
is uploaded from the first level of its chain within the cap. Compression is
lossy, which makes it the choice for colour, normal and detail maps; `rgba8`
holds a texture exactly, which is what pixel art and a mask whose exact codes
matter need.
`effort` is how long the encode searches for each block. `fastest` tries one
candidate for an opaque block, `fast` takes about 6 times as long and
`thorough` about 17 times, each landing closer to the source texels. What the
GPU holds is the same size at every effort, and so is the time a load takes.
`uastcRdo` trades accuracy for a payload that compresses smaller: `none`
encodes each block as close to the source as the encoder finds, and `low`,
`medium` and `high` let a block move further from the source when a nearby
block's bytes would let the payload compress better. What the GPU holds is the
same size either way; what moves is the size of `data.ztex` on disk and over
the network, and the encode time, which a pass more than doubles.
The encode runs on the CPU when the texture is created or a setting that
decides its texels changes. A call from a task that can yield cooks off the
frame and the task waits for it, so the world keeps running: a 1024x1024
texture with its mip chain takes about a second on a desktop CPU.
A texture whose settings say `rgba8` is never compressed.
## Reading an asset from its ref
`texRef:settings()` reads the container's settings with every schema default
filled in and `texRef:setSettings({ maxDimension = 1024 })` patches them;
`texRef:load()` decodes the payload's pixels into the CPU store for per-pixel
access, `texRef:handle()` puts the texture on the GPU under the asset's own
guid, and `texRef:preview()` renders it as a swatch.
`texRef:info()` reads the compiled `data.ztex` payload's header: the format it
was compiled to — also as its `compression` and `colorSpace` — the
`textureType`, `alpha` and `alphaIsTransparency` it was encoded under, its width and height, its mip and array-layer counts, the
sampler recorded in it (`filter`, `wrapU`, `wrapV`, `anisotropy`), and, for an animated sequence, each layer's
display time and the length of one pass:
```lua
local info = texRef:info()
print(info.width, info.height, info.format, info.mipCount)
```
It answers off that payload's own header, so a compiled texture reports how big
it is for a header read where `load()` decodes its pixels and `handle()` uploads
them. A container whose primary is a source image answers `(nil, err)`, and
`load()` reads its size. The `compression` and `colorSpace` it reports are what
the payload holds; the settings table holds what the encode was asked for.
## In the Inspector
Selecting a texture opens its **Texture** section right below the asset's
identity, then its Settings. The section draws the encoded texture through the
texture's own sampler, so a `filter = "nearest"` texture reads crisp at every
zoom. The zoom row offers **Fit** and each integer zoom (1×, 2×, 4×, 8×) whose
drawn size fits 512 px; a 128 px texture offers up to 4×, and one wider than
512 px offers Fit alone. Below the view, the section lists every field of the
`data.ztex` header (`texRef:info()`) — size, format, mip levels, filter, layers,
and a sequence's frames and duration — and the GPU memory the encoded mip chain
takes across its layers. The filter and the other encode settings change in
the Settings section; the Texture section reports the re-encoded result.
▲ 0↑ born
# `.font` asset type
A `.font` is a first-class font asset: font-file bytes (TTF / OTF / WOFF /
WOFF2) converted into a guid-anchored asset via
`asset.create("font", name, { bytes = … })`.
A font is a **general CPU resource** usable by any text surface. The
`onRegister` hook calls the engine `font.register(name, bytes)` primitive,
which installs the parsed face into **both** the 2D/3D text system (`text.*`)
and the egui UI text system, under the family name the font file itself
reports. A registration is what puts the face in the database; whether
`fontFamily = "<name>"` selects it is the separate question of whether the
shaper matches that name, and `font.reconcile()` answers it per family:
```lua
for _, f in ipairs(font.reconcile()) do
if not f.selectable then
print(f.family .. " is registered but shapes as " .. f.shapedWith)
end
end
```
`text.face(handle)` gives the same answer for one label. A family the shaper's
database holds is selectable under its own weights and for the glyphs its faces
cover: a row that reads `familyCoveredNoGlyph` names a family the shaper picks
whose faces have no glyph for the text asked of it. `topics/text` covers reading
a label that came out in the wrong face.
Fonts are CPU resources (not GPU), so the primitive lives in the `font.*`
namespace, the CPU analogue of `renderer.texture.create`.
## Layout
```
<name>.font/
data.zfnt the baked payload `asset.create("font", ...)` writes
face.json optional: family group, role, aliases, on-demand entries
```
## Editing a font's files
A write to a file the registration reads (`face.json`, `data.zfnt`, or a raw
font file in the folder) installs the font again from what the folder now
holds. The names the new `face.json` gives it (its group and each alias) select
it on every text surface from then on, and a name the earlier descriptor gave it
and the new one does not stops selecting it; a changed `group` moves the face to
that group. Text laid out under a name the edit decides is laid out again, and a
`Text3D` label draws it again on the next frame. For an on-demand face the
same holds for the font and coverage files its entries name: the entries are
declared again from their coverage files, and an entry already installed is
registered again from its font file with the names the descriptor now gives.
## On-demand faces
A face whose file is large and whose script most sessions never draw is
declared by the codepoints it covers instead of being read at boot. Its
`face.json` lists the font files of the folder under `onDemand`, each beside a
coverage file:
```
NotoSansCJK.font/
face.json { group, role, aliases, license, onDemand = [ { name, file, coverage }, ... ] }
kana.otf a font file (TTF / OTF)
kana.coverage the codepoints it maps: "3000-30ff,ff01-ff5e,..."
...
OFL.txt the license the files are distributed under
```
`onRegister` reads the coverage files and calls `font.declare(name, coverage,
{ file, family, role, aliases })` for each entry; no font file is read. The first
time a label or a `text.*` object meets a character an entry covers and no
installed face draws, the engine requests that entry: its file is read without
blocking the frame and registered with `font.register`, and from then on every
text surface draws the character from it. `font.onDemand()` reports each entry
as `declared`, `requested`, `installed` or `failed`, and `font.prefetch(text)`
requests the entries a text needs before it is drawn.
`font.coverage(bytes)` writes a coverage file from a font file's character map.
Splitting one large face into several entries (by script, and by how common a
character is) means a session reads only the part its text reaches: the
built-in `NotoSansCJK.font` is cut that way, and a page of Japanese reads its
kana and common-ideograph entries alone.
Every registered face, on-demand or not, is a per-glyph fallback for every
family the UI holds: a character the selected family has no glyph for is drawn
from the first face that has one.
## In the Inspector
Selecting a font opens its **Font** section: the family the face registers under, its role and aliases from `face.json`, whether the text system has the family registered and, when it holds it, how many faces and which weights. Below the facts, a sample line is drawn in the family at 14, 20 and 32 px.
▲ 0↑ born
# Shader (asset type)
A shader is a WGSL program that runs on the GPU. Materials reference shaders and
supply property values; renderable entities reference materials, not shaders
directly. This README is the canonical reference for writing **every** kind of
shader — it covers all domains, the surface authoring contract, the material
property schema, the engine library functions you can call, and how a shader
registers and compiles.
> New to how shaders, materials, meshes, and textures relate as assets and live
> GPU resources? Read `core/resource-model` first.
## Folder shape
A `.shader` is a folder asset:
```
<name>.shader/
shader.wgsl # the WGSL body (required)
properties.yaml # the material property schema (optional; surface shaders)
README.md # a one-paragraph description of THIS shader (required)
.metadata # asset metadata
```
Each file carries a committed `.meta` sidecar pinning its stable guid. The
identity is `<name>` (the `.shader` suffix strips).
A shader's material properties are exactly the ones its `properties.yaml`
declares: a shader with no `properties.yaml` declares none, and the engine
reads no property from the WGSL body. A `.wgsl` file standing on its own,
outside a `.shader` folder, is WGSL source rather than a shader: a material
draws with the `.shader` folder that holds it as `shader.wgsl`. Materials and other assets
reference a shader by **AssetRef** (a typed, guid-backed handle), never by a raw
string.
## Domains
The first line of `shader.wgsl` declares the domain:
```wgsl
// @domain: surface
```
| Domain | What it is | You write |
|---|---|---|
| `surface` | A material on a mesh (the common case). | `fragment()` (+ optional `vertex()`/`vertex_clip()`). The engine generates everything else. |
| `post-process` | A full-screen pass over the rendered frame, in the space it declares (bloom, vignette, grade). | `fragment(in: PostInput) -> vec4` only. The engine generates the fullscreen `vs_main`/`fs_main` and the group(1) interface from `properties.yaml`. |
| `screen` | A standalone full-screen visual (UI background, procedural). | `fragment(in: ScreenInput) -> vec4` only. The engine generates the fullscreen `vs_main`/`fs_main` and the group(1) interface from `properties.yaml`. |
| `sky` | The sky background (a fullscreen material). | `fragment(in: SkyInput) -> vec4` only. The engine generates the fullscreen `vs_main`/`fs_main` and the group(1) interface from `properties.yaml`. |
| `compute` | GPU data processing, no rasterization. | An `@compute @workgroup_size(...)` entry over storage buffers; you write the `@group/@binding` lines yourself. For compute work, the `.computeShader` assetType (see below) generates them for you. |
`// @domain:` resolves the names above and `postprocess` / `screenspace` / `ui`
as further spellings of `post-process` and `screen`. A body carrying no
`@domain:` line is `surface`. `shaderRef:domain()` answers the domain a body
declares, a further spelling as the domain it names. Any other name is refused where the shader
registers, with an error naming the spellings that resolve, so a shader reaches
the renderer only under a domain the compiler scaffolds.
Particles are drawn and simulated through these domains like anything else: the
particle package's `particle_sprite` and `particle_mesh` are `surface` shaders,
and its simulation, culling and sorting steps are `.computeShader` assets.
## Texture coordinate space
A shader that samples a texture through a coordinate it builds itself declares
which space that coordinate lives in, on a line beside `@domain:`:
```wgsl
// @domain: surface
// @uv_space: world
```
| Value | The coordinate | A material's `uv_scale` means |
|---|---|---|
| `mesh` (the default) | The mesh's own UV attribute, `input.uv`. | How many times the texture repeats across a surface. |
| `world` | The fragment's world position, as a triplanar projection samples. | How many times the texture repeats per world unit, so a larger surface carries more. |
Texture residency reads the declaration to work out how finely a frame samples
each texture: a material laying eight copies of its texture across a surface
samples each copy at an eighth of the surface's size, and the mip level that
serves it is three levels coarser than the surface's own footprint implies.
Leaving the line out puts the shader in mesh space, which is what sampling
`input.uv` is. `renderer.textureStreaming().textures` reads back the span each
texture was measured at and the level that span asked for.
**Compute shaders have their own assetType.** A `.computeShader/` folder gives compute the same zero-scaffolding contract surface/sky/post/screen get: you write only `@compute fn main` (+ helper fns/structs, **no `@group/@binding` lines**) and declare bindings in a sibling `bindings.yaml` (storage buffers, textures, samplers, storage textures, plus a `params:` uniform); the engine generates the `@group(0)/@binding(N)` declarations. Dispatch it through a resolved reference (`asset.resolve("name", "computeShader"):dispatch(...)`). See the `.computeShader` assetType README and `man compute`.
**Surface**, **sky**, **post-process**, and **screen** shaders all get
engine-generated scaffolding — you write only `fragment()` and declare properties in
`properties.yaml`; the engine generates the entry points and the group(1) material
interface. (Each domain's `fragment` takes a domain-specific input instead of
`FragmentData`: `SkyInput`, `PostInput`, or `ScreenInput` — see "Sky shaders" and
"Post-process & screen shaders" below.) Compute shaders get the same treatment via
the separate `.computeShader` assetType (write only `@compute fn main`, declare
bindings in `bindings.yaml`). The `compute` domain here writes its own
entry points directly — see `man compute`.
## Surface shaders
A surface shader writes **only** the parts that are unique to it. The engine owns
all the scaffolding — the vertex stage, every render-mode entry point
(`vs_main` / `fs_forward` / `fs_gbuffer`), group(0) (camera, transforms, lights,
shadows, and the material rows), and the material interface generated from
`properties.yaml` (`material` and your texture slots). Adding a render path never
touches your shader.
Inside `fragment()` you have access to exactly four things — nothing is hidden:
1. **`input: FragmentData`** — the interpolated per-fragment surface data the
engine hands you: world position, world normal, uv, view direction, vertex
colour, and more. Every field is in the `FragmentData` table below.
2. **`material.*` and your textures** — the material inputs **you** declare in
`properties.yaml`. The mapping is mechanical: a scalar property named `foo`
becomes `material.foo`; a texture property named `bar` becomes `bar` (the
texture) and `bar_sampler` (its sampler). `material` has **exactly** the fields
your `properties.yaml` declares and no others — to add an input, add a property.
List what a shader exposes with `shaderRef:getProperties()` or
`asset.inspect("<name>")`.
3. **Engine state** — the scene's camera, lights, time, and shadows that the
renderer binds for every surface shader. `uniforms.*` (sun direction/colour,
ambient, `camera_pos`, `time`) and the punctual lights (`zero_point_light(i)`)
are always in scope; the shadow functions come from `#include "@builtin::shaderModules.shadows"`. This is how
you light a surface **without** the prebuilt PBR — see "Engine state" and
"Writing your own shading" below.
4. **Library functions** you `#include` (see "Library includes").
A shader and its property schema go together. The two files below are a complete
surface shader:
`properties.yaml` — declares the material inputs:
```yaml
properties:
- { name: base_color, type: color, default: [1, 1, 1, 1] }
- { name: roughness, type: float, default: 0.5, min: 0, max: 1 }
- { name: metallic, type: float, default: 0.0, min: 0, max: 1 }
- { name: base_color_texture, type: texture, default: white }
```
`shader.wgsl`; those declarations are read as `material.base_color`,
`material.roughness`, `material.metallic`, and the `base_color_texture` /
`base_color_texture_sampler` pair:
```wgsl
// @domain: surface
#include "@builtin::shaderModules.pbr_shading" // the PBR helpers — see "Library includes"
fn fragment(input: FragmentData) -> vec4<f32> {
var s = zero_pbr_default_surface(input);
s.albedo = material.base_color.rgb
* textureSample(base_color_texture, base_color_texture_sampler, input.uv).rgb;
s.perceptual_roughness = material.roughness;
s.metallic = material.metallic;
return vec4<f32>(zero_pbr_shade(input, s), material.base_color.a);
}
```
`fragment()` returns the final colour. The engine forces no lighting model on
you — flat, toon, PS1, dithered, or PBR are all just different `fragment()`
bodies. Calling `zero_pbr_shade` is opt-in.
### The two surface bodies, and what each gives up
A surface shader exposes one of two entry points, and which one decides how
much of the shader the engine can see:
| Body | The engine gets | PBR debug passes read |
|---|---|---|
| `fn fragment(input: FragmentData) -> vec4<f32>` | the finished pixel, and the `PbrSurface` too when the body hands one to `zero_pbr_shade` | the real channel where a surface was handed over; magenta, the "no data here" marker, where the body shades by its own math |
| `fn surface(input: FragmentData) -> PbrSurface` | the surface (albedo, roughness, metallic, emissive, normal, occlusion), which the engine then lights | the real channel |
Both are first-class: `fragment()` is what you want when the shader owns its
own look, `surface()` when you want engine lighting. `capture { pass = "albedo" }`
(and `roughness` / `metallic` / `ao` / `emissive`) answers about your material
for either body wherever the engine has the surface — writing the same
`PbrSurface` behind `surface()` or behind `fragment()` + `zero_pbr_shade` gives
one answer per channel. A body that lights itself by its own math draws those
channels magenta. `pbr.shader` exposes `surface()`.
`capture { pass = "normal" }` is a geometry read rather than a material one, so
it answers for every shader: the normal of the surface where one was handed
over, and the interpolated normal of the triangle where none was. It is the same
normal the frame's normal buffer carries, which is what a screen-space pass
reading `@scene.normal` sees.
`shaderRef:shadingModel()` reports which one a shader ended up with —
`"engine-lit"` for `surface()`, `"self-shading"` for `fragment()`.
### Optional vertex hooks
```wgsl
// Displace geometry in OBJECT space. Runs on skinned meshes too (it sees the
// already-skinned, morph-applied local position). Return the modified vertex.
// The `v` parameter is immutable (WGSL function params are), so copy it into a
// `var` first and modify that.
fn vertex(v: VertexData) -> VertexData {
var out = v;
out.position += out.normal * sin(out.time) * 0.1;
return out;
}
// Post-transform vertex hook (runs AFTER the engine's world/skin/projection). Two
// things you can do here, separately or together:
// * move `clip_position` — vertex jitter, PS1 snapping, render-resolution lowering;
// * write `color` — PER-VERTEX (Gouraud) shading: light the vertex from its
// world_normal + the light context and the engine forwards the result into the
// interpolated `vertex_color` the fragment reads.
fn vertex_clip(v: ClipVertex) -> ClipVertex {
var out = v;
// Gouraud: Lambert sun + ambient, evaluated once per vertex (flat-banded shading).
let n_dot_l = max(dot(v.world_normal, -normalize(v.light_direction)), 0.0);
let lit = v.ambient_color * v.ambient_intensity + v.light_color * v.light_intensity * n_dot_l;
out.color = vec4<f32>(v.color.rgb * lit, v.color.a);
out.clip_position = round(out.clip_position * 80.0) / 80.0; // PS1 snap
return out;
}
```
#### Declaring how far a hook moves geometry: `vertexBoundsDilation`
Static geometry is culled per cluster against its bounds as authored. A hook that moves vertices — wind sway, an outline hull, a vertex wave — can move them outside those bounds, and the cull then drops clusters whose displaced triangles are still on screen. Declare the farthest the hook moves any vertex, in world units, as a `vertexBoundsDilation` float property, with a default that covers the shader's own properties across their ranges:
```yaml
properties:
- { name: sway, type: float, default: 0.5, min: 0, max: 4 }
# 0.4 x sway, at sway's largest value.
- { name: vertexBoundsDilation, type: float, default: 1.6, min: 0, max: 100 }
```
Each material sets its own value — a material that sways less can set less, one moved farther sets more — and a material that sets none takes the default. The cull adds the material's value to every bound in its frustum and occlusion tests (not to the level-of-detail cut). A shader with a hook and no `vertexBoundsDilation` property is named once in the engine log when it compiles; its geometry is culled against its undisplaced bounds.
### Structs the engine passes you
`FragmentData` (input to `fragment()`):
| Field | Type | Meaning |
|---|---|---|
| `world_position` | `vec3<f32>` | fragment world-space position |
| `world_normal` | `vec3<f32>` | interpolated geometric normal (normalized) |
| `world_tangent` | `vec4<f32>` | xyz = world tangent, w = handedness (mikktspace) |
| `uv` | `vec2<f32>` | primary texture coordinates (perspective-correct) |
| `affine_uv` | `vec2<f32>` | the same UVs interpolated **without** perspective correction — the PS1 texture "swim". Always available; sample with it instead of `uv` for affine texture mapping |
| `vertex_color` | `vec4<f32>` | interpolated per-vertex colour (white if unset; carries per-vertex lighting if `vertex_clip()` wrote it — see Gouraud below) |
| `view_position` | `vec3<f32>` | camera world position |
| `view_direction` | `vec3<f32>` | normalized world-space direction to the camera |
| `light_direction` | `vec3<f32>` | directional (sun) travel direction, unit length; surface→sun is `-light_direction`. Where no sun reaches the camera it points straight down |
| `light_color` | `vec3<f32>` | sun colour |
| `light_intensity` | `f32` | sun intensity; 0 where no sun reaches the camera |
| `ambient_color` | `vec3<f32>` | ambient colour |
| `ambient_intensity` | `f32` | ambient intensity |
| `frag_coord` | `vec4<f32>` | framebuffer pixel coords (xy) + depth (z) — for screen-space dither |
| `receives_shadows` | `f32` | per-instance: > 0.5 if this surface receives shadows |
| `material_index` | `u32` | per-instance: the index standing for the material this surface draws with — the number `renderer.materialIdentity()` maps to a name |
| `shader_data` | `array<vec4<f32>, 4>` | per-instance: this entity's own four `vec4` lanes — its `ecs.ShaderData` component's `lane0`..`lane3`, or what `renderer.instanceData.set` put in them (zero until something does). A procedural graph's elements set them through the `data0`..`data3` attributes |
| `front_facing` | `bool` | true for front faces (double-sided shading) |
| `time` | `f32` | gameplay seconds: advances with the world at `engine.timeScale` and stands still while gameplay is paused (edit mode starts paused), so an effect slows and stops with the world |
`shader_data` is the per-instance channel, and it is what lets ONE material serve
many entities that differ in a value — a dissolve at its own progress per subject,
a hit flash at its own age, a fill level per instance. The alternative is a
material per entity, which is a separate pipeline binding and a separate row in
the material table for every one of them.
```wgsl
const DISSOLVE_LANE: u32 = 0u;
fn fragment(input: FragmentData) -> vec4<f32> {
let progress = input.shader_data[DISSOLVE_LANE].x;
...
}
```
```lua
-- one material on twenty entities, each at its own progress
renderer.instanceData.set(subject, 0, progress)
renderer.instanceData.clear(subject) -- back to zero lanes
```
The engine attaches no meaning to a lane: a shader and whatever writes it agree
what each one carries. Name the indices in the module that writes them.
`renderer.instanceData.laneCount()` is how many there are. The same lanes reach a
hardware-instanced population through `renderer.mesh.drawInstanced`'s
`instanceDataBuffer`, on slots no entity owns.
`VertexData` (in/out of `vertex()`): `position` (object-local, model-scale + morph
applied), `normal`, `uv`, `color`, `time`.
`ClipVertex` (in/out of `vertex_clip()`) — the post-transform vertex. Read context:
`clip_position`, `world_position`, `world_normal`, `uv`, `view_position`,
`view_direction`, `light_direction`, `light_color`, `light_intensity`,
`ambient_color`, `ambient_intensity`, `time`, `view_proj` (the view-projection of
the frame being evaluated), and where the vertex stands on its instance:
`local_position` (its position in the mesh, after `vertex()` and before the
instance's world transform), `object_origin` (the world position of the instance's
origin), `object_scale` (the world length of the instance's x, y and z axes) and
`object_axes` (`mat3x3<f32>`, the world direction of those axes, unit length) —
what a camera-facing card, a shard grown about its own base, or anything placed
per instance is built from. **Writable:** `clip_position` (snap /
jitter) and `color` (write a lit colour here for per-vertex Gouraud shading; the
engine forwards it to the interpolated `vertex_color` — left untouched, the mesh's
own vertex colour passes through).
`vertex()` and `vertex_clip()` run in the vertex stage, which reads the frame
uniforms (`uniforms.*`, and the `ClipVertex` members built from them), the scene
buffer (`zero_object_data`, `zero_object_visuals`, the node transforms), the
material interface and a `read` buffer property. The point, spot and area lights
(`zero_point_light`, `zero_punctual_projection`), the shadow maps, the
environment and light probes, the feature textures, the scene depth, colour and
reflection copies and a `@group(2)` buffer the shader declares `read_write` are
visible to the fragment stage alone: read them in `fragment()` or `surface()`. A
shader whose vertex hook reaches one fails to compile, and the error names the
stage, the binding and the calls that reach it.
### Each hook runs twice: this frame and the one before it
A vertex hook decides where the surface IS, so it also decides where the surface
WAS. The engine runs every hook a shader declares a second time, over the same
vertex standing one frame back — the previous-frame position stream, with
`time` = the clock that frame was drawn at and (for `vertex_clip()`)
`view_proj` = the view-projection it was drawn with. The motion vector the
renderer publishes is taken between the two results, so a displacement that
moves with time reports the velocity of the displaced surface, and a
displacement that does not move reports the motion of the transform alone
because the identical offset appears on both evaluations and cancels.
What that asks of a hook:
- **Take the frame from the argument.** `v.time` is the clock of the evaluation
being run; `uniforms.time` is always this frame's. A displacement built on
`uniforms.time` is identical on both evaluations, so its motion vector cancels
to exactly zero. The same holds for `v.view_proj` against
`uniforms.view_proj`: a clip hook that rebuilds its position from world space
through `uniforms.view_proj` reprojects the previous frame with this frame's
camera, and reports no camera motion at all. The wind helpers come in both
shapes for this reason — `zero_wind_sway_at_time(world_pos, stiffness, v.time)`
inside a hook, `zero_wind_sway(world_pos, stiffness)` outside one.
- **Anchor a wrap to `uniforms.time`.** A displacement that jumps discontinuously
— a falling drop that restarts at the top, a looping offset — differences its
jump as velocity on whichever frame it wraps. Compute the wrap COUNT from
`uniforms.time` and the position within the cycle from `v.time`, and both
evaluations share the same cycle, so the surface reports the speed it fell at.
`@builtin::shaders.rain_material` and `snow_material` do exactly that.
The per-vertex attributes (`world_normal`, `uv`, `color`) and the lighting and
camera-position context are this frame's on both evaluations, so a hook that
displaces along its normal reports the motion of the transform plus this frame's
displacement.
## Engine state — point lights & shadows
The common lighting state — the sun (`input.light_direction` / `light_color` /
`light_intensity`), the ambient (`input.ambient_color` / `ambient_intensity`), the
camera (`input.view_position` / `view_direction`), and `input.time` — is already on
`FragmentData`, so most shaders never touch the raw engine state. The one thing not
on `FragmentData` is the punctual-light array (its length varies), which is always
in scope:
| Expression | Type | Meaning |
|---|---|---|
| `uniforms.point_light_count` | `u32` | number of active point/spot lights |
| `zero_point_light(i)` | `PointLightData` | one point or spot light (fields below) |
| `zero_punctual_projection(i)` | `PunctualProjection` | a spot or area light's projection: `view_proj`, the frustum its cookie and shadow are placed by, and `cookie_layer` |
`PointLightData`: `position`, `radius`, `color`, `intensity`, `direction`,
`cone_outer_cos`, `cone_inner_cos`, `kind` (0 = point, 1 = spot).
Shadows come from `#include "@builtin::shaderModules.shadows"`:
- `compute_directional_shadow_factor(world_pos, normal, light_dir) -> f32` — 1.0 lit, 0.0 fully shadowed.
- `compute_spot_shadow_factor(...)` / `compute_point_shadow_factor(...)` / `compute_area_shadow_factor(...)` — one shadow shape each.
- `compute_punctual_shadow_factor(light_index, world_pos, normal) -> f32` — the row's own shape, read off its `kind`, for a loop that walks the punctual lights and does not want the switch.
- `zshadow_light_reach(light_index, world_pos, normal, object_channels) -> f32` — how much direct light that row puts on a surface on those light channels: the channel mask, falloff, cone or rect orientation, and the cosine, times intensity. Zero exactly where the light contributes nothing.
## Writing your own shading
The engine forces no lighting model — `zero_pbr_shade` is just one option. You can
light a surface yourself from the state above. Here is a complete custom-lit
(Lambert diffuse) surface with no PBR:
```wgsl
// @domain: surface
#include "@builtin::shaderModules.shadows" // exposes compute_directional_shadow_factor
fn fragment(input: FragmentData) -> vec4<f32> {
let n = normalize(input.world_normal);
let albedo = material.base_color.rgb;
// Directional (sun) light: Lambert N·L, attenuated by its shadow.
let sun = normalize(-input.light_direction); // surface -> sun
let sun_shadow = compute_directional_shadow_factor(input.world_position, n, input.light_direction);
var lit = input.light_color * input.light_intensity * max(dot(n, sun), 0.0) * sun_shadow;
// Point + spot lights (the punctual-light array, varying length).
for (var i = 0u; i < uniforms.point_light_count; i = i + 1u) {
let pl = zero_point_light(i);
let to_light = pl.position - input.world_position;
let dist = length(to_light);
let l = to_light / max(dist, 1e-4);
let atten = max(1.0 - dist / max(pl.radius, 1e-4), 0.0);
lit += pl.color * pl.intensity * max(dot(n, l), 0.0) * atten;
}
// Ambient fill.
lit += input.ambient_color * input.ambient_intensity;
return vec4<f32>(albedo * lit, material.base_color.a);
}
```
- **Unlit:** skip the lights and return your colour directly.
- **Toon / PS1 / custom:** start from the loop above and band, quantize, or warp
it however you like.
- **PBR:** build a `PbrSurface` and call `zero_pbr_shade` (see "Library includes")
instead of writing the BRDF by hand.
## Material properties — `properties.yaml`
A surface shader declares its material properties in a sibling `properties.yaml`.
The engine GENERATES the entire group(1) interface from it — the uniform struct
(`material.<name>`) plus a texture/sampler pair per declared texture. You write
no `@group`/`@binding` lines.
```yaml
properties:
- { name: base_color, type: color, default: [1, 1, 1, 1] }
- { name: roughness, type: float, default: 0.5, min: 0, max: 1 }
- { name: metallic, type: float, default: 0.0, min: 0, max: 1 }
- { name: emissive, type: color, default: [0, 0, 0, 1] }
- { name: base_color_texture, type: texture, default: white }
- { name: normal_texture, type: texture, default: flat_normal }
```
| `type` | WGSL field type | Notes |
|---|---|---|
| `float` | `f32` | optional `min`/`max` for editor sliders |
| `float2`/`float3`/`float4` | `vec2/3/4<f32>` | |
| `color` | `vec4<f32>` | sRGB colour, default is `[r,g,b,a]` |
| `int` | `i32` | |
| `bool` | `u32` | |
| `texture` | `texture_2d<f32>` + `sampler` | `default:` is a built-in texture name (`white`, `black`, `flat_normal`) |
In the shader you read scalars as `material.<name>` and sample textures as
`textureSample(<name>, <name>_sampler, uv)`. A texture slot with no bound texture
samples its default, so a shader can always sample — no presence flag needed.
`vertex()` reads the same material interface, so a height or deformation map
drives displacement from the slot it was bound to. A vertex stage has no implicit
derivatives, so sample it at an explicit mip there:
`textureSampleLevel(<name>, <name>_sampler, uv, 0.0)`.
## Sky shaders
A `// @domain: sky` shader draws the sky as a fullscreen pass. You write **only**
`fn fragment(in: SkyInput) -> vec4<f32>` — the colour along the view ray for each
pixel — and declare tunables in `properties.yaml` exactly like a surface shader
(read as `material.<name>`, textures via `textureSample(<name>, <name>_sampler, uv)`).
The engine generates the fullscreen `vs_main`/`fs_main`, the group(1) interface,
and reconstructs the world-space ray; there is no separate sky renderer — a sky
material runs through the same material pipeline, just on a fullscreen triangle.
A sky pipeline binds group(0) and the group(1) interface alone, so a sky that
reads a `buffer` property or any other `@group(2)` binding fails to compile, and
the error names the binding and the calls that reach it.
`SkyInput` (the only argument) carries:
| Field | Type | Meaning |
|---|---|---|
| `ray_dir` | `vec3<f32>` | normalized world-space ray through this pixel |
| `ndc` | `vec2<f32>` | screen position in NDC `[-1, 1]` |
| `sun_direction` | `vec3<f32>` | unit vector TOWARD the sun (from the directional light) |
| `camera_pos` | `vec3<f32>` | world-space camera position |
| `time` | `f32` | gameplay seconds: advances with the world at `engine.timeScale` and stands still while gameplay is paused (edit mode starts paused), so an effect slows and stops with the world |
```wgsl
// @domain: sky
fn fragment(in: SkyInput) -> vec4<f32> {
let up = max(in.ray_dir.y, 0.0);
let col = mix(material.horizon_color.rgb, material.zenith_color.rgb, up);
return vec4<f32>(col * material.exposure, 1.0);
}
```
Day/night comes from `sun_direction.y` (the directional light's elevation), so a
sky shader never needs a time-of-day uniform — orient the light and the sky
follows. The builtin sky shaders (`procedural_sky`, `solid_sky`, `cubemap_sky`,
`equirect_sky`) are all authored this way; author a custom sky material and point
a `Skybox` component at it.
## Post-process & screen shaders
`// @domain: post-process` and `// @domain: screen` shaders are authored exactly
like surface and sky — you write **only** `fn fragment(in: PostInput) -> vec4<f32>`
(post) or `fn fragment(in: ScreenInput) -> vec4<f32>` (screen) and declare tunables
in `properties.yaml` (read as `material.<name>`, textures via
`textureSample(<name>, <name>_sampler, uv)`). The engine generates the fullscreen
`vs_main`/`fs_main` and the group(1) material interface. **There are no
`uniforms.params[i]` slots** — that convention is retired; every input is a named
property.
A **post-process** shader's `properties.yaml` carries one further, top-level key
beside `properties:` — `space`, either `light` (the default) or `display`:
```yaml
space: display
properties:
- { name: amount, type: float, default: 1.0, min: 0, max: 1 }
```
`light` is the scene's linear radiance — what every surface, the sky and every
`afterLighting` feature write, unbounded above 1.0. `display` is the code values
the `__display__` pass leaves behind after applying exposure and its tone curve
— the 0-1 range a colourist's tool and a dithered or quantized effect expect.
`postprocess.add`'s `opts.space` overrides what the asset declares for one
registration. The chain converts at both ends of an effect that runs on a
segment in the other space, so a light-space vignette still darkens a frame
graded after `__display__`, and a lookup table built from a `.cube` export
always sees the code values it was authored against, wherever the chain
runs it.
`space` also decides which layer a registration that names none lands on: a
`light` shader grades `scene`, the segment holding the scene's own light, and a
`display` shader grades `all`, the picture after `__display__` with the UI on
it. So `postprocess.add("glow", asset.resolve("~.shaders.bloom", "shader"))`
thresholds highlights that still stand above 1.0, and a lookup table registered
the same way reads the code values it was built for. `opts.layer` names one
explicitly — `"all"` is what an effect that must grade the interface says.
A **post-process** shader's `properties.yaml` carries a further, optional
top-level key — `passes:`, the pass list the frame draws through, in order:
```yaml
space: light
passes:
- { name: bright, entry: bright, scale: 0.5, inputs: [scene] }
- { name: down1, entry: downsample, scale: 0.25, inputs: [bright] }
- { name: up0, entry: upsample, scale: 0.5, inputs: [down1, bright] }
- { name: composite, entry: composite, inputs: [scene, up0] }
```
Every pass's entry — `fn <entry>(in: PostInput) -> vec4<f32>` — lives in the
one `shader.wgsl` the assetType recompiles on. `scale` is the fraction of the
view a pass draws at; the LAST pass takes none, since it draws at the view's
size and writes the chain's output. `inputs` names what a pass reads —
`"scene"`, the chain's input for this effect, or the name of an EARLIER pass —
and `inputs[0]` is what `zero_sample_scene(uv)` reads inside that pass's body;
every input, `inputs[0]` included, is also reachable by index through
`zero_pass_input(i, uv)` and `zero_pass_input_size(i)`, below. The one place
the two reads of `inputs[0]` disagree: in an effect whose declared space
differs from the segment it draws on, `zero_sample_scene` converts a `scene`
input into that space while `zero_pass_input(0, uv)` hands back the segment's
own, unconverted value. A shader declaring no `passes:` is one pass whose
entry is `fragment`, and its generated WGSL is byte-identical to what it is
without the key.
A pass that is not the last writes a scratch target the effect owns — in the
space the shader declared, view-sized times its own `scale` — allocated on the
first draw and replaced whenever the size changes. `engine.params`, below,
carries the drawing pass's own index, its scale, and its first input's texel
size, so one shader body serves every level of a chain built this way.
The worked example is the bloom: `@builtin::shaders.post.bloom` declares
twelve passes — a threshold pass, five downsamples, five tent upsamples and a
composite — out of four entries, since the five downsamples share one and the
five upsamples share another, each level reading the one below it through
`zero_pass_input` and telling itself apart by `engine.params`. Its own README
walks the declaration.
A declaration is refused at compile time — reported through `compileStatus()`,
`postprocess.add`'s second return value, or a live effect's
`status()[i].error` — for: an entry `passes:` names that `shader.wgsl` does
not define; an input naming a later pass, or naming neither `scene` nor an
earlier pass; a `scale` outside `(0, 1]`; a `scale` declared on the last pass;
a `buffer` property (group(2) carries the pass inputs); or a shader whose
group(0) plus declared texture slots plus widest pass's inputs ask for more
than the 16 FRAGMENT sampled textures one shader stage has.
A **post-process** shader runs as a full-screen pass over the already-rendered
frame. Its `fragment` argument and framework helpers:
`PostInput`:
| Field | Type | Meaning |
|---|---|---|
| `frag_coord` | `vec4<f32>` | framebuffer pixel coords (`@builtin(position)`) |
| `uv` | `vec2<f32>` | screen UV, `(0,0)` top-left … `(1,1)` bottom-right |
Framework helpers (always in scope — no include needed):
| Call | Returns | Meaning |
|---|---|---|
| `zero_sample_scene(uv)` | `vec4<f32>` | the rendered scene colour at `uv`, converted into the space this effect declared (`zero_scene_light` is the same call, for content that reads it by that name) |
| `zero_scene_depth(uv: vec2<f32>)` | `f32` | scene depth at `uv` — a screen UV here, where the surface domain's `zero::scene_depth` module takes a pixel coordinate under the same name (exact texel — an R32Float copy). Reversed: 1.0 is the near plane, 0.0 the far plane and empty sky, so a nearer surface reads GREATER and `d > 0.0` means "a surface was drawn here" |
| `zero_linear_depth(d, near, far)` | `f32` | that sample as a distance in metres — the one inversion of the reversed range |
| `zero_scene_motion(uv)` | `vec2<f32>` | per-pixel screen-space motion vector at `uv` |
| `zero_pass_input(i, uv)` | `vec4<f32>` | the pass's declared input `i`, at `uv`. Input 0 is what `zero_sample_scene` reads |
| `zero_pass_input_size(i)` | `vec2<f32>` | that input's size in texels, so one entry serves every level of a chain |
Engine data uniform `engine` (a `PostEngineData`):
| Field | Type | Meaning |
|---|---|---|
| `resolution` | `vec4<f32>` | `x`=width, `y`=height, `z`=1/width, `w`=1/height |
| `time` | `vec4<f32>` | `x`=elapsed seconds, `y`=dt, `z`=frame index |
| `view_proj` | `mat4x4<f32>` | this frame's view-projection |
| `prev_view_proj` | `mat4x4<f32>` | last frame's view-projection (motion) |
| `inv_view_proj` | `mat4x4<f32>` | inverse view-projection (world-space reconstruction) |
| `params` | `vec4<f32>` | `x`=this pass's own index, `y`=the scale it draws at, `zw`=1/size of its first input — for a pass the post-process chain draws. A pass a `renderFeature` fragment pass draws carries that feature's `spec.params` here instead; a shader declaring `passes:` is always drawn by the chain, so the two never meet in one shader. |
The engine fills all of it **per render target**. A registered effect runs over
the live viewport's frame AND over every offscreen one — a `capture` from a
world-space station or orbiting an entity, a capture of a named camera, a
render-to-texture camera — and in each of those, `view_proj`, `prev_view_proj`
and `inv_view_proj` are the camera THAT render was drawn from, while
`resolution` is that target's own pixel size. So a pass reconstructing world
space from `zero_scene_depth(uv)` reconstructs against the station and lens the
capture asked for, and an offscreen frame is a picture of the effect as it
stands at that station. (A render feature's passes reach an offscreen render the
same way, against the same camera — the `renderFeature` assetType README carries
that half, including `@frame.camera`, its own per-target camera input.)
An offscreen render keeps no view history of its own, so `prev_view_proj` there
holds that same matrix rather than the frame before it, and a pass taking camera
motion from the two reads none. On the viewport the pair is a frame apart, so
camera motion is the one reading a capture answers differently from the screen.
`postProcessing = false` on a capture is the one control that takes the chain
off the frame it returns; it names the camera property the ephemeral capture
camera is built with, so it scopes to that one capture. What a capture opts out
of is the registered effects; `__display__` still runs on it, so the exposure
and tone curve that turn the scene's light into a picture are on what a capture
returns.
A diagnostic capture (`albedo`, `normal`, `motion_vectors`, …) bypasses
`__display__` itself, since a channel like that is a reading of the scene
rather than a graded picture of it. The environment-cube capture bypasses it
too, and for the same reason: a probe face carries radiance into the cube
array, and `__display__` would fit that radiance to a display's code values —
a picture of it rather than the value the PBR indirect term multiplies into
its lighting.
A complete post-process effect — a vignette via `fragment()` + `properties.yaml`:
`properties.yaml`:
```yaml
properties:
- { name: intensity, type: float, default: 0.5, min: 0, max: 1 }
- { name: radius, type: float, default: 0.75, min: 0, max: 1 }
- { name: softness, type: float, default: 0.45, min: 0, max: 1 }
```
`shader.wgsl`:
```wgsl
// @domain: post-process
fn fragment(in: PostInput) -> vec4<f32> {
let color = zero_sample_scene(in.uv);
let intensity = max(material.intensity, 0.01);
let radius = max(material.radius, 0.01);
let softness = max(material.softness, 0.01);
let center = in.uv - vec2<f32>(0.5);
let dist = length(center);
let vignette = smoothstep(radius, radius - softness, dist);
let darkened = color.rgb * mix(vec3<f32>(1.0), vec3<f32>(vignette), intensity);
return vec4<f32>(darkened, color.a);
}
```
The asset above **defines** the effect. It enters the frame when something
registers it:
```luau
postprocess.add("vignette", asset.resolve("~.shaders.vignette", "shader"), { priority = 200 })
```
The registration names no layer, so the effect grades the layer its `space`
belongs to: this vignette reads light, so it runs on `scene`, over the picture
before the UI reaches it. Registering it is the one step that makes it render,
so a world's chain holds exactly the passes its content asked for. Editing the shader body afterwards
recompiles every effect registered from it, in place — the pass keeps its enabled
state, priority, layer, tuned property values and bound textures, and draws the
new pixels on the next frame. `postprocess.remove(name)` takes it back out of the
chain and leaves the asset alone.
Drive the named properties at runtime with `postprocess.setProperty(name, prop,
value)` and bind texture properties with `postprocess.setTexture(name, prop, path)`
(both keyed by the **property name**, never a slot index — see the post-process
section of `man shaders` / the `postprocess.*` API).
Read them back with `postprocess.describe(name)`: it carries the chain state
`postprocess.status()` lists for that effect, the `properties` schema it was
registered with, and the `values` each of those properties currently holds. Before
the effect exists, `shaderRef:getProperties()` on the asset names the same schema,
so what a pass takes is answerable either side of the registration.
A **screen** shader is a standalone full-screen visual (UI background, procedural
pattern). `ScreenInput` mirrors `PostInput` (`frag_coord` + `uv`); there is no scene
to sample, so it has no `zero_sample_scene` helpers. Its `engine` uniform is a
`ScreenEngineData`:
| Field | Type | Meaning |
|---|---|---|
| `time` | `f32` | gameplay seconds: advances with the world at `engine.timeScale` and stands still while gameplay is paused (edit mode starts paused), so an effect slows and stops with the world |
| `scroll` | `f32` | UI scroll offset |
| `resolution` | `vec2<f32>` | render-target size in pixels |
| `mouse` | `vec2<f32>` | mouse position |
| `velocity` | `f32` | scroll velocity |
```wgsl
// @domain: screen
fn fragment(in: ScreenInput) -> vec4<f32> {
let t = engine.time * 0.5;
let col = vec3<f32>(
sin(in.uv.x * 3.14159 + t) * 0.5 + 0.5,
sin(in.uv.y * 3.14159 + t * 0.7) * 0.5 + 0.5,
material.brightness
);
return vec4<f32>(col, 1.0);
}
```
## Library includes
A library module is a **`.shaderModule` asset**: it only EXPOSES functions and
structs and does nothing on its own (exactly like a Unity `.cginc`). You must
call what you include. Include one by its identity, the same string you would
hand `asset.resolve`:
| Include | Exposes |
|---|---|
| `@builtin::shaderModules.pbr_shading` | `struct PbrSurface { albedo, normal (world), metallic, perceptual_roughness, emissive, occlusion }`; `zero_pbr_default_surface(FragmentData) -> PbrSurface`; `zero_pbr_shade(FragmentData, PbrSurface) -> vec3<f32>` (the full forward PBR lighting) |
| `@builtin::shaderModules.shadows` | `compute_directional_shadow_factor(...)`, `compute_spot_shadow_factor(...)`, `compute_point_shadow_factor(...)`, `compute_area_shadow_factor(...)`, `compute_punctual_shadow_factor(...)`, `zshadow_light_reach(...)` |
| `@builtin::shaderModules.engine_bindings` | the group(0) declarations (camera, transforms, lights, shadow maps) — surface shaders get this implicitly; include it directly only for advanced raw access |
| `@builtin::shaderModules.shadow_bias` | `zshadow_offset_receiver(...)`, `zshadow_directional_depth_bias(...)` and the rest of the bias math, as pure functions |
| `@builtin::shaderModules.texturing` | `ztex_stochastic(...)` (tile without a visible repeat grid), `ztex_blend_normals(...)` (reoriented normal blending) |
| `@builtin::shaderModules.depth` | `zero_linear_depth(depth, near, far)` — the one reading of a stored depth |
| `@builtin::shaderModules.gbuffer_material` (`zero::gbuffer_material`) | `struct ZeroGbufferMaterial { base_color, metallic, perceptual_roughness }`; `zero_gbuffer_unpack_material(texel: vec4<u32>) -> ZeroGbufferMaterial`; `zero_gbuffer_material(t: texture_2d<u32>, coord: vec2<i32>) -> ZeroGbufferMaterial`, `zero_gbuffer_base_color`/`zero_gbuffer_metallic`/`zero_gbuffer_roughness(t, coord)` — the one unpacking of `@scene.material`'s `Rgba8Uint` texel into what a surface reflects |
| `zero::scene_depth` | `zero_scene_depth(px: vec2<f32>)`, `zero_scene_depth_uv(uv: vec2<f32>)`, `zero_scene_depth_texel(px: vec2<i32>)`, `zero_scene_depth_size()`, `zero_scene_depth_is_sky(d: f32)`, `zero_scene_world_position(px: vec2<f32>)`, `zero_scene_world_normal(px: vec2<f32>)`, `zero_scene_depth_fade(px: vec2<f32>, world_position: vec3<f32>, fade_distance: f32)` — every `px` is a pixel coordinate in two components, so a surface fragment passes `input.frag_coord.xy`, not the `vec4<f32>` `FragmentData.frag_coord` holds. The depth already drawn UNDER a surface fragment, and the world position and geometric normal of the surface standing there. What a decal projector box, a soft particle, a forcefield or water intersection foam shades against. Surface domain; the depth belongs to the view being drawn |
| `zero::scene_color` | `zero_scene_color(px: vec2<f32>)`, `zero_scene_color_uv(uv: vec2<f32>)`, `zero_scene_color_texel(px: vec2<i32>)`, `zero_scene_color_size()`, `zero_scene_color_refract(px: vec2<f32>, offset_px: vec2<f32>)`, `zero_scene_color_refract_uv(px: vec2<f32>, offset_fraction: vec2<f32>)` — every `px` is a pixel coordinate in two components, so a surface fragment passes `input.frag_coord.xy`, not the `vec4<f32>` `FragmentData.frag_coord` holds. The scene colour already composed BEHIND a surface fragment, and the displaced read that bends it. What heat haze, a refraction shockwave, a cloaking or shield warp, glass and ice shade with. Surface domain; the colour belongs to the view being drawn, and is the lit opaque scene as it stood immediately before the transparent pass — so a `blend`-queue surface reads this frame's, an opaque one reads the previous frame's, and two overlapping blended surfaces both read the same picture and do not bend each other |
`asset.list("shaderModule")` enumerates the `.shaderModule` assets in this
engine, builtin and world-authored alike, and each one's README says what it
exposes. Those are the `@builtin::shaderModules.*` rows above; a module you
write is included exactly the same way, by its own identity or by the
`~.name` form.
The `zero::`-named rows come from the renderer itself. Their WGSL is baked into
the engine binary and registered under those names before any asset exists, so
they answer an `#include` in every engine and appear in no asset index. A shader
that includes one records no dependency on it: the engine reading the shader is
the engine that already holds the module.
Two spellings reach the same `.shaderModule`: a module's declared alias
answers as well as its identity, so a shader written against `zero::pbr`
compiles and pins the asset that name resolves to. Prefer the identity: it
says which asset travels with the content.
Whichever punctuation an include is written with — `#include "name"`,
`#include 'name'` or `#include <name>` — names the same module and records the
same dependency.
The engine's own scaffolding — `zero::surface`, `zero::sky`, `zero::post`,
`zero::screen` — is injected by the compiler around the matching domain and is
never written by hand. The surface framework puts `safe_normalize`,
`mat3_from_mat4` and `apply_normal_map(FragmentData, tangent_normal)` in scope
for every surface shader.
## Shader features — compile-time variants
A **property** is runtime uniform data (a colour, a float, a texture): set per
material, read as `material.<name>`, no recompile. A **feature** decides which
code is in the program at all. Use a property when the value changes at runtime,
a feature when it changes what the shader is.
Declare features in `properties.yaml`, alongside the properties:
```yaml
properties:
- { name: base_color, type: color, default: [1, 1, 1, 1] }
features:
- { name: PARALLAX, default: false, description: silhouette parallax march }
- { name: DETAIL, default: true, description: detail-normal overlay }
```
Guard the code each feature owns with `#ifdef`:
```wgsl
fn surface(input: FragmentData) -> PbrSurface {
var s = zero_pbr_default_surface(input);
#ifdef PARALLAX
s.normal = parallax_normal(input); // absent when PARALLAX is off
#else
s.normal = input.world_normal;
#endif
return s;
}
```
`#ifdef` / `#ifndef` / `#else` / `#endif`, one per line, and they nest. The
directives resolve before the WGSL reaches the compiler, so the program holds
exactly the lines its feature set selected and every other line is an empty one,
which keeps a compiler message on the line number the file shows. An unbalanced
block is a compile error naming its line.
A material is what chooses a feature set, so `features:` shapes the shaders a
material draws with — the surface, sky and screen-space domains.
A material names the features it wants in its `mat.yaml`:
```yaml
name: worn_rock
shader: "@builtin::shaders.pbr_layered"
features:
PARALLAX: true # on, over the shader's default
DETAIL: false # off, against the shader's default
```
or through `asset.create`:
```lua
asset.create("material", "worn_rock", {
shader = "@builtin::shaders.pbr_layered",
features = { PARALLAX = true, DETAIL = false },
})
```
### A property that switches a feature on
A feature can also be switched on by a material's property values. List the
properties under the feature's `properties:`, and the feature is on for every
material whose value for one of them stands away from that property's declared
default:
```yaml
properties:
- { name: clearcoat, type: range, default: 0.0, min: 0, max: 1 }
features:
- { name: CLEARCOAT, properties: [clearcoat], description: the coat lobe }
```
```wgsl
#ifdef CLEARCOAT
s.clearcoat = material.clearcoat;
#endif
```
A material at `clearcoat: 0` draws a program holding none of the coat's code; one
that sets `clearcoat: 0.6`, in its `mat.yaml`, through `asset.create` or with
`renderer.material.setProperty`, draws the `CLEARCOAT` variant, and setting it
back to 0 returns it to the base program. The property stays a runtime value the
program reads; the feature decides only whether the code reading it is there.
This is how a shader with costly optional parts keeps every material that leaves
them at rest on the cheap program without each material having to name a
feature. A material naming the feature off under `features:` keeps it off
whatever the property holds. Write the guarded code so that, with the feature
off, the shader computes what it would with the property at its default: the
material sees one look either way, and pays for the part only where it shows.
Each distinct set of switched-on features is a program to compile the first time
a material reaches it. A material keeps the programs its values switched it onto
compiled while it lives, so a property crossing back and forth reuses them; when
the shader's variant budget is spent, the ones no material draws with are given
back first. A `properties:` entry is a value property (`float`, `range`, `int`,
`bool`, `color`, `float2` to `float4`) the shader declares; one naming no declared
property, or a texture or buffer, is refused by name.
The engine compiles one program per feature set its materials ask for, named
`<shader>#<features>`. Materials sharing a set share that program. A feature the
shader does not declare selects nothing, so a stray name cannot make the engine
compile anything, and the compile names both halves of a mismatch: a guard on a
name `features:` does not declare, and a declared name no guard tests.
**Permutations are the cost.** A shader with n features can express 2^n
programs, each its own compile and pipeline. One shader holds at most 32
variants; a material asking for a set past that draws with the shader's default
features and the engine logs which shader ran out. Read what a scene is
spending:
```lua
for _, s in renderer.shaderVariants() do
print(s.shader, #s.variants, "of", s.budget)
print(renderer.variantSource(s.base.program)) -- what it ships as
for _, v in s.variants do
print(v.program, renderer.variantSource(v.program)) -- what one asked for
end
end
```
A shader registers under every key it answers to — its identity, its guid, each
alias — and each key is a row. A variant lands on the row for the key the
material named, which for authored content is the guid, so take a program name
from the report rather than composing one.
## Materials reference shaders
A `.material` applies a shader to entities. Its `mat.yaml` references the shader by
its identity string and supplies property values:
```yaml
name: my_material
shader: "@builtin::shaders.pbr"
render:
blend: alphaBlend
cull: back
depth_write: false
properties:
roughness: 0.5
base_color: [1.0, 1.0, 1.0, 1.0]
textures:
base_color_texture: { ref: "default:white" }
```
`shader:` is a plain identity string, and a colour is an array. A key's
value is the rest of its line, so every value above stands alone on one.
### The render block
Render behaviour (the blend equation, alpha
cutout, which faces are drawn) lives in the `render:` block and drives the
pipeline, not the fragment. These are the keys that reach the pipeline from both
authoring routes, and the values each one takes:
| Key | Values | Sets |
|-----|--------|------|
| `blend` | `opaque` `transparent` `alpha` `alphaBlend` `additive` `add` `cutout` `alphaCutout` `premultiplied` `premultipliedAlpha` `multiply` | the equation the fragment composites with |
| `cull` | `back` `front` `none` `off` `disabled` | which faces the rasterizer keeps |
| `depth_write` | `true` `false` | whether the draw writes depth |
| `depth_compare` | `less` `lessEqual` `lequal` `equal` `greater` `greaterEqual` `gequal` `always` `never` | the test a fragment's depth passes |
| `topology` | `triangleList` `lineList` `lines` `line` | what consecutive vertices assemble into |
| `queue` | a whole number: `2000` opaque, `3000` transparent, `4000` overlay | the draw order of the material; a number inside a queue orders within it |
`blend` is also spelled `type`, and the two name one key. Its equations, over
the fragment the shader returned (`src`) and the value already in the target
(`dst`): `opaque` is `src`; `transparent`, `alpha` and `alphaBlend` are
`src * src.a + dst * (1 - src.a)`; `additive` and `add` are `src + dst`, the
mode a glow, a light shaft or a spark card is written for; `premultiplied` and
`premultipliedAlpha` are `src + dst * (1 - src.a)`, taking a source that already
carries its own coverage; `multiply` is `src * dst`, which is how a soot, grime
or shadow card darkens what stands behind it. `cutout` and `alphaCutout` keep
the fragment or discard it on its alpha and go on writing depth.
The same block is what `asset.create`'s `render` argument takes, key for key and
value for value, so one material named in Luau and the same material standing on
disk ask for their pipeline with the same words. `depth_write` is the one key
whose value takes the shape of its route: the word `false` on a `mat.yaml` line,
the boolean `false` in the Luau table.
```lua
asset.create("material", "heat_plume", {
shader = "@builtin::shaders.pbr",
render = { blend = "alphaBlend", cull = "none", depth_write = false },
})
```
Where a value outside a key's set is reported is what the two routes differ on.
`asset.create` refuses the call on any of these keys and names the set it
measured the value against, or for `queue`, that it takes a number. A
`mat.yaml` standing on disk registers either way, each key falling back to a
value of its own: `blend` to `opaque`, `cull` to `back`, `depth_write` to
`true`, `depth_compare` to `less`, `topology` to `triangleList` and `queue` to
`2000`. The engine log carries the blend fallback, naming the value and
the set it was measured against, so a `mat.yaml` value is worth putting through
`asset.create` once — that is where every key answers with its set.
## Registration & hot reload
A `.shader` is compiled by the asset system, not by the renderer: writing
`shader.wgsl` or `properties.yaml` fires the `.shader` assetType's `onChange`,
which compiles it: the body + the property schema are read, the group(1)
interface and (for surface) every render-mode entry point are generated,
`#include`s are expanded, the result is **validated**, and the shader is
registered. Validation
happens at this compile step — the whole point of "compile". A shader with invalid
WGSL (an unknown field, a type mismatch, an undefined function) **fails to compile**
with the real reason — readable directly from Luau via
`shaderRef:compileStatus()` (`{ status = "compiled" | "failed" | "pending",
error?, pending? }`), and also in the author log and `asset.inspect`; it is **not**
registered. This fires on create and on every edit, so saving a `.wgsl` recompiles
with no world reload; the change is visible on the next frame. Compilation is
async, so a `compileStatus()` of `"pending"` immediately after a write means check
again next frame.
A surface shader is also checked for the material resolve, the pass that
shades each pixel of the visibility buffer through the shader's own code. That
check reaches its verdict a few frames after the raster compile. Until it
does, `compileStatus()` reads `"pending"` with `pending` saying so, and the
shader already draws. A shader the check refuses reads `"failed"`, and `error`
names the construct the resolve cannot evaluate and where it sits; its pixels
resolve through the error shader until it changes.
The stage the compile checks comes from one of two places: the hook the engine
scaffolds entry points around (`fragment()`, or `surface()` for an engine-lit
surface), or the `vs_main` / `fs_main` pair the body writes itself. The pair is
a pair: a body that writes its own entry points owes both stages. A body
carrying neither route — a hook whose name is a letter off, say, or half a pair
— holds no stage the rasterizer can enter, so the compile **refuses it** and
names both the entry point it looked for and what the body carries, in the same
`compileStatus()` error every other compile failure reads back through.
### Reading a compile error against the text it is about
What the compiler reads is not the file you wrote. The engine generates the
group(1) material interface from `properties.yaml`, wraps the body in its
domain's framework and every entry point that domain needs, expands every
`#include`, and resolves every `#ifdef` — so a 100-line body reaches naga as
tens of thousands of lines. **Every line number in a compile error, and every
`[N]` handle index naga prints where it has no name to print, is a position in
that composed module.** A fault at module scope (`no definition in scope for
identifier: 'zero_scene_depth'`) reads fine without it; a fault inside a
function body does not.
`renderer.compiledSource(name)` returns that composed module, and
`shaderRef:compiledWgsl()` returns it for a ref you already hold. Both answer
for a shader that declares no features, needs no material, entity or draw, and —
the case they exist for — **answer for a shader whose compile failed**, holding
the text the compiler was refused over for as long as `compileStatus()` reports
that failure. `renderer.compiledShaders()` lists every name they answer for.
```lua
local ref = asset.resolve("myShader", "shader")
local status = ref:compileStatus()
if status.status == "failed" then
print(status.error) -- the line, the caret, and the cause
local wgsl = ref:compiledWgsl() -- the text those positions are in
local line = tonumber(string.match(status.error, "wgsl:(%d+):"))
print(string.split(wgsl, "\n")[line])
end
```
The composed text carries every `zero::*` module the shader included, so the
signature a call was rejected against is in the same text as the rejection —
which is the whole of what a "wrong argument type" error needs to be actionable.
A material bound to a broken (or missing) shader renders with the **magenta error
shader** — never invisibly — so a shader problem is impossible to miss on screen as
well as in the log. **Fixing the shader and re-saving recovers it live**, with no
engine restart: the recompile clears the failed-shader state and the object returns
to normal on the next frame.
## Discovery
- `asset.list("shader")` — every registered shader.
- `asset.inspect("<name>")` — domain, status, source, parse errors, this README.
- `shaderRef:compileStatus()` — did it compile? `{ status, error?, pending? }`, no
log needed. `"pending"` with `pending` set means the material resolve check
is still reaching its verdict; a resolve refusal reads as `"failed"`.
- `shaderRef:compiledWgsl()` / `renderer.compiledSource("<name>")` — the WGSL the
compiler received, which is what a compile error's positions index into.
Answers for a failed compile too.
- `renderer.compiledShaders()` — every name that call answers for.
- `shaderRef:getProperties()` — the parsed `properties.yaml` schema.
- `shaderRef:listMaterialsUsing()` — materials bound to this shader.
## Common pitfalls
- **A compile error shows up in one of three ways at the draw.** A shader that
never compiled has no pipeline, so the renderer binds the magenta placeholder.
A shader that compiled once and was then edited into brokenness KEEPS the
pipeline that build produced — the surface goes on drawing the last good
program, pixel for pixel, and looks entirely intact while the source is broken.
A geometry-pass-only shader has no magenta tell either, since it is not on a
forward mesh. `shaderRef:compileStatus()` gives the status + error text (as
does `asset.inspect("<name>")`), and `renderer.drawDiagnostics()` names which
renderables are in which of the three states — a row carries `substituted` for
the placeholder, `stale` for a program that no longer compiles, and the cause
either way.
- **A compile error's line numbers are lines of the composed module, not of your
file.** `renderer.compiledSource("<name>")` / `shaderRef:compiledWgsl()` is
that module; resolve the position against it rather than counting lines in
`shader.wgsl`. The `[N]` a message prints in place of a name is a naga handle
index — it moves with the size of the body and is the same value for two
unrelated faults in one file, so it is neither a line number nor a way to tell
two failures apart.
- **Don't declare group(0) or group(1) in a surface shader.** The engine owns
group(0) and generates group(1) from `properties.yaml`. Re-declaring them is the
fastest way to a layout mismatch.
- **`#include` does nothing by itself** — it only exposes functions; you must call
them.
- **Renaming a property** changes the generated uniform field; materials setting the
old name silently fall back to its default.
- **Compute shaders don't render** — verify output via a buffer readback or a
debug pass. Author them as a `.computeShader` (its own assetType), not a
`.shader` — see `man compute`.
## Related types
- `.material` — the required wrapper that applies a shader to entities.
- `.style` — UI cascade, not GPU shaders.
## In the Inspector
Selecting a shader opens its **Shader** section: its domain and shading model (and, for a post-process shader, the space its `fragment()` reads and returns), whether it compiled (and the compiler's error when it did not), a table of its declared properties (type, default, range), its post-process passes when it has any, the bindings it declares, and openers for its source. The **Materials** section (collapsed at first) has a Find materials action that lists, as links, every material drawing with the shader; it scans every material, so it runs only when asked.
▲ -1↑ born
backing path · assetTypes/mesh.assetType
Problems
Everything affecting this asset right now: its own problems, anything wrong inside it, and problems on its direct dependencies.
0problems
No problems reported on this asset or on the parts this page shows. Parts it leaves out are not counted here.
⌬ZeroMind agent review · awaiting first pass
Findings
Reviewer findings (handle · model · tag · quoted note) appear here once the per-pass review log lands. Today only the rolled-up
agent_score is exposed.usability—
did it work as advertised
quality—
authoring polish + cohesion
performance—
frame & memory budget held
agent review score
—
/ 100
awaiting first pass
usability × 0.40
+ quality × 0.35
+ performance × 0.25
± compat factor
+ quality × 0.35
+ performance × 0.25
± compat factor
Usability ratings
Did the part work as advertised when consumers tried to drop it in. Separate from upvotes: those are taste; this is "did it function".
—%no reports yet
Sign in to report whether this part worked for you.
Discussion
Scoped to this part · feeds back into the world's score.
0comments
Sign in to post.sign in
No comments yet. Be the first.