A #[node_macro::node] function returning a struct tagged with
field is a named output connector (title-cased from the field name,
renamed with #[name("...")], described by its doc comment). By default
the node has a hidden primary output carrying the whole struct with the
fields as secondary outputs; marking at most one field #[primary] makes
that field the primary output instead.
The macro generates one hidden extractor proto node per field plus a
registration keyed by the struct's TypeId. The Graphene preprocessor
recognizes nodes returning a registered struct and substitutes them, in
the transient runtime copy of the network only, with a generated network
exporting each field through its extractor. The destructuring machinery
therefore never appears when drilling into a node, in copied clipboard
content, or in saved documents. When a Memoize implementation is
registered for the struct type, the struct is computed once and shared
across all outputs rather than re-evaluated per output.
The editor derives output counts, names, and types for such nodes from
the registry. The old hand-authored "Split Vec2" and "Split Channels"
wrapper-network definitions are replaced by multi-output split_vec2 and
split_channels proto nodes, with document migrations that keep existing
wires valid since the output indices are unchanged.
The "Position on Path" and "Tangent on Path" nodes are combined into a
single multi-output "Evaluate Path" node whose primary output is the
position and whose secondary output is the tangent angle. A migration
converts old instances, forwarding the shared inputs and remapping the
tangent nodes' downstream connections to the new tangent output index.
The now-redundant "Extract XY" node is removed (its role is subsumed by
Split Vec2's destructuring), and "Extract Channel" becomes a plain helper
used by Split Channels rather than a standalone node.
* graster-nodes: change `Table<Color>` params to `Color` where only one value is used
* Re-add support for Color and Option<Color>
* Add warning when a default value isn't parsed
---------
Co-authored-by: hypercube <0hypercube@gmail.com>
* Prototype document network level into node insertion
* Implement Convert trait / node for places we can't use Into
* Add isize/usize and i128/u128 implementations for Convert trait
* Factor out substitutions into preprocessor crate
* Simplify layer node further
* Code review
* Mark preprocessed networks as generated
* Revert changes to layer node definition
* Skip generated flag for serialization
* Don't expand for tests
* Code review
---------
Co-authored-by: Keavon Chambers <keavon@keavon.com>
