Files
oak-editor/crates/oakundo
Mike-Solar 2db1615453 feat(oakplugin): wire OpenFX plugins into the node graph and renderer
- oaknode: dynamic node factory registration, PluginNode value model
  pushing PluginJobPayload, traverser texture passthrough for texture
  inputs, type-stamped RefBox::get_checked.
- oakrender: PluginExecutor dependency-inversion slot; eval resolves
  and executes plugin jobs, purple frame on failure.
- oakplugin: node_factory with full OFX param -> node input
  translation (15 types, color semantics heuristic, combo ordering,
  secret/ui_group/ui_page, clip inputs), plugin instance registry,
  render executor + duplicator installation, progress reporter and
  active-viewer provider injection points, U8/U16/F16 input
  conversion with NaN scrubbing, in-place output frame writeback fix.
- gl_bridge.rs documents the wgpu<->GL interop spike: Metal-first on
  macOS rules out wgpu-hal GL interop; offscreen GL context deferred.

End-to-end tests cover registration, param translation, CPU render
pixel assertions, identity passthrough and NaN fallback.
2026-08-18 17:15:13 +08:00
..

oakundo Rust crate

Status: implemented. Ports the C++ oakundo module (src/undo/src) to Rust behind its frozen C ABI (include/undo/*.h). Template follows crates/oakplugin.

Scope

Replaces the C++ oakundo module (src/undo/src): undoable commands and the undo/redo history stack. Public contract: include/undo/*.h (3 headers: error.h, undocommand.h, undostack.h) — frozen, implemented verbatim by src/ffi.rs.

Architectural decisions

  1. Vtable-command pattern is the centerpiece. In C++ other modules subclass olive::UndoCommand (redo()/undo() overrides) and plug themselves in polymorphically. Rust has no inheritance, so the C ABI already models exactly this with OakUndoCommandVtable { redo, undo, free_fn } plus a caller-owned userdata pointer. The safe layer's [undocommand::CommandKind] is the direct analog: either a caller-defined vtable command (function pointers + userdata) or a [undocommand::MultiUndoCommand] composite. Domain logic dispatches on the vtable the same way the C++ virtual dispatch does.
  2. Modified-state callbacks are intentionally not part of the C ABI. The C++ UndoCommand::redo_and_set_modified pair records/restores a project dirty flag via std::function accessors. The public headers expose none of this; the stack drives state via done_ on the safe type instead, and the flag callbacks are left as a documented future extension.
  3. UndoStack state machine is modeled directly on the C++: two deques — commands_ (done, oldest at front) and undone_commands_ (most-recently-undone at front); push clears any redoable tail, executes redo, and drops the oldest when the cap (200) is exceeded; jump clamps and walks via undo/redo. The fresh stack holds a single "New/Open Project" empty command so can_undo is false at the bottom (per undostack.cpp).
  4. No merge semantics. include/undo/*.h and src/undo/src/* define no merge_with/can_merge; commands are never coalesced. Tests reflect this (no merge tests).

Layout

src/
  lib.rs            crate doc + module map
  error.rs          error codes (include/undo/error.h)
  handle.rs         refcounted-handle scaffolding (OAKUNDO_ABI_VERSION=1)
  undocommand.rs    UndoCommand / vtable command / MultiUndoCommand
  undostack.rs      UndoStack + empty bottom command
  ffi.rs            export layer (one submodule per public header)
tests/              contract tests per module

error.h exports macros only and is folded into ffi.rs's preamble (no own submodule), matching the codec crate convention.

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.