M0a of the render-pipeline plan (docs/zh/plans/render-pipeline-threads.md): - oak-node: every payload push site (58 across footage.rs, plugin.rs and the nodes/* effects) now boxes the Job enum instead of the raw payload. The enum gains CacheJob with a CacheJobPayload (path + time + fallback value, the C++ cachejob.h shape), plus safe as_* accessors and unsafe probe helpers beside job_ref. - oak-render: RenderEvalHooks::resolve is one loop over the table — a single get_checked::<Job> probe per texture value, a match dispatch to process_footage/shader/plugin/color_transform/cache, and recursive resolution of the job boxes embedded in a payload's inputs (depth-capped, cycle-guarded) — replacing the four sequential full-table scans (resolve_*_jobs, deleted). - The disk frame cache is real: frameio.rs implements a minimal self-describing F32 container (magic/version/dims/format/timestamp + payload, tmp-write + atomic rename, full header validation on load) because the OIIO bridge is a stub and EXR is unavailable in this build; process_cache_job genuinely reads the file before falling back to the job's (already resolved) fallback value. - Tests: CacheJob roundtrip (save -> resolve -> pixel equality), missing-file fallback, nested cache-job-through-shader resolution, plus four frameio container tests. 2330 passed, 0 failed across the workspace.
oaknode Rust crate (implementation)
Status: all FFI headers implemented. Phase 1 (core engine: graph arena, values, keyframes, project, factory, ~55 FFI exports), Phase 2 (sequence/track/block/footage/colormanager + traverser + serializer, the folder/group/keyframe/dragger FFI families, the undo/XML bridges with test stubs, and the contract tests) and Phase 3 (the multicam grid family and the deferred bridge exports: markers/work-area/frame cache accessors, viewer params, sequence/footage stream params via the videoparams/audioparams C ABIs, and the colormanager compliant transform) are complete;
cargo test --features test-stubsis green (84 tests, 1 ignored byte-exact golden). The remainingtodo!()s are the concrete node-type behaviors undersrc/nodes/(registered in the factory, bodies deferred) — the multicam node behavior and the effect/generator nodes. The crate template (FFI discipline, testing layers) followscrates/oakplugin/README.md.
Scope
Replaces the C++ oaknode module (src/node/src, ~40k lines):
the node graph engine, project/folder/sequence/track/block hierarchy,
footage, color manager, keyframes, evaluation (traverser), project
serialization, and the undo bridge.
Public contract: include/node/*.h (14 headers, ~280 functions) —
frozen, implemented verbatim by src/ffi.rs.
Key architectural decisions (C++ → Rust mapping)
- Inheritance → arena + trait objects. The C++ design is deep
inheritance (
Node→ViewerOutput/Track/Block/… and ~50 effect nodes). Rust: a slab-allocatedGrapharena ofNodeEntry { core: NodeCore, behavior: Box<dyn NodeBehavior> }, addressed by generationalNodeId. No reference cycles exist by construction (edges are IDs, not pointers). - Cross-module inheritance disappears. C++
RenderProcessor : NodeTraverser(render subclassing a node class) becomes a plain evaluation API:traverser::evaluate(...) -> NodeValueTableis a function, and oakrender supplies backend hooks via a trait (RenderHooks) instead of overriding virtuals. - Value system.
olive::Variant/type-erasure becomes a closedNodeValueenum (value.rs). C ABI marshalling lives only inffi.rs. - Undo. Commands are created through the oakundo C ABI
(
bridge::undo); the C++UndoCommandsubclass hierarchy becomes vtable commands whose userdata is a Rust closure. - Serialization. XML read/write goes through the oak_core C ABI
(
bridge::common) until oak_core itself is rewritten. - Threading. The C++ code relied on Qt's event thread +
called_on_owner_thread()assertions. Rust replaces this withMutex<Graph>interior mutability plus explicit&mut Graphphases for structural edits; the threading contract is documented per function.
Layout
COVERAGE.md maps every method of the C++ olive::Node (260
declaration lines, ~150 unique methods) to its Rust home — trait /
core / graph / ops / bridge / drop-with-reason. Review that first.
src/
lib.rs crate doc + module map
error.rs error codes (mirrors include/node/error.h)
handle.rs refcounted-handle scaffolding (same pattern as plugin)
value.rs NodeValue / NodeValueTable / ValueHint
id.rs NodeId, generational arena ids
node.rs NodeCore + NodeBehavior trait (the virtual surface)
graph.rs Graph arena, edges, topological order
input.rs Input descriptors, flags, array inputs, hints
keyframe.rs NodeKeyframe + track interpolation
project.rs Project, settings, folder tree
sequence.rs Sequence (ViewerOutput equivalent)
track.rs Track, TrackList
block.rs Block/ClipBlock/GapBlock/TransitionBlock
footage.rs Footage (probe via oakcodec C ABI)
colormanager.rs ColorManager (OCIO via oakrender C ABI for now)
traverser.rs Evaluation engine (iterative, hook-based)
serializer.rs XML project load/save (bridge::common)
factory.rs Node type registry (id -> constructor)
nodes/ The concrete built-in node types
bridge/ C ABI imports: common.rs, undo.rs, render.rs, codec.rs
ffi.rs include/node/*.h export layer
tests/ contract + golden tests (see README test section)
Hard rules for the implementer
- Every
extern "C"body goes throughhandle::guard*; no panic crosses FFI. Graphis the only owner of nodes; the public API never hands out references into the arena, onlyNodeId-carrying handles.- Behavior parity with C++ is proven by the C ABI test-suite
(
src/node/tests, unchanged) plus the golden tests intests/. - Where C++ behavior is genuinely load-bearing but ugly (e.g.
Blocklength-change side effects onTrack), port the behavior, not the aesthetics; leave a// CPP-PARITY:comment with the C++ file:line.
Dependency policy
Prefer mature third-party crates (MIT/Apache-2.0/BSD, GPL-compatible) over hand-rolling; register each addition (name + reason) here. Large existing C++ libraries (OTIO, OCIO, OIIO, FFmpeg) are NEVER rewritten — they are consumed through their C ABI / bridge layers.