Crate engawa_lisp 

Tatara-lisp authoring layer for engawa render graphs.

Operators write effect declarations in a .tlisp file:

(defmaterial scanlines
  (shader (inline "..."))
  (bindings
    (binding 0 uniform "frame")
    (binding 1 texture "scene")
    (binding 2 sampler "scene-sampler")))

(defresource swap external)
(defresource scene (texture 800 600))
(defresource post (texture 800 600))

(defgraph mado-pipeline
  (input swap)
  (output post)
  (node clear-scene (kind clear) (output scene))
  (node scanlines-pass (kind fullscreen-effect)
    (material scanlines)
    (input scene)
    (output post)))

…and this crate parses + lowers to engawa::RenderGraph. Pairs with shikumi’s notify watcher for live hot-reload — save the .tlisp, the graph rebuilds, the dispatcher picks up the new topology without a restart.

Why a custom lisp parser

The full tatara-lisp macro engine is heavy. For engawa’s IR — which is intentionally small (8 types) — a minimal sexpr parser + typed lowering does the job in ~600 LOC. When the tatara-lisp ecosystem stabilises (caixa-author lands, the macro engine becomes reusable across consumers), engawa-lisp migrates to lean on it.

Today: round-trip-tested sexpr parser, every parse error carries operator-friendly context (line:col + the form being parsed), 30+ unit tests.

Re-exports

pub use lower::lower_to_graph;

pub use lower::LowerError;

pub use parse::ParseError;

pub use parse::Span;

pub use sexpr::Sexpr;

pub use sexpr::SexprKind;

Modules

lower

Typed lowering — parsed sexprs → engawa::RenderGraph.

parse

Minimal s-expression tokenizer + parser.

sexpr

Typed s-expression tree — the parser’s output, the lowering step’s input. Pure data; no parser state, no lifetimes.

Enums

EngawaLispError

Functions

parse_and_lower

One-call helper: parse a .tlisp source string, lower to an engawa::RenderGraph. Returns the graph ready for compile; the caller chains .compile() to get a CompiledGraph for dispatch.