oakstorage: the sqlite URI parse tests used /tmp/lib.db, which is not absolute on Windows, so parse_target's is_absolute check rejected it. Pick the absolute path per platform (C:/tmp/lib.db on Windows). ci (Linux): wrap the test step in a 1500 s watchdog — a deadlocked test prints nothing and never fails; on timeout the watchdog dumps every test/worker process's thread stacks with gdb and then kills the suite. (One such hang already ate a run; the previous green run needed ~4 min.) ci+cd: Swatinem/rust-cache gains cache-on-failure everywhere, so a red run still saves its compile cache (the actions/cache FFmpeg cache already saves in its post phase regardless of outcome).
oakstorage Rust crate — project persistence
Status: implemented (file backends). Manual: docs/zh/plans/riir/M10-oakstorage.md.
Scope
Project persistence — the single module that knows where projects come
from and where they are saved to. Backends are pluggable via a manual
vtable; shipping in this pass: ove-xml (the XML project format) and
otio (the .otio / .fcpxml interchange, via the native oakotio
crate). The database backend (PostgreSQL + SQLite, SeaORM) is a
declared stub for a later proxy — not registered, todo!() bodies.
Consumers never branch on backend.
Architectural decisions
- URI dispatch, not file paths. Every entry point takes a URI:
file:///…proj.ove/file:///…proj.otio/oakdb://…. Bare paths are normalized tofile://. The core resolves scheme + backendcan_handlearbitration (M10 §2.3). - Manual vtable backends (
backend.rsStorageBackendtrait = the M10 C vtable's Rust shape). The public C ABI vtable (oakstorage_backend_register) accepts foreign (C-side) backends — the database-swap interface proof — and in-crate backends implement the Rust trait directly. - The graph (de)serialization itself stays in oaknode — every
backend calls the oaknode serializer (
oaknode::serializer::load/save) throughbridge::node(direct Rust calls, single-lib unification) to fetch/rebuild the in-memory graph; backends own framing: container bytes, schema, versioning (TOO_OLD/TOO_NEW/ UNKNOWN_VERSION), sessions.OAKSTORAGE_SAVE_COMPRESSis accepted but not implemented (the oaknode serializer emits plain XML only). - Database backend shares one logical schema across PostgreSQL and SQLite via SeaORM: a private current-thread tokio runtime drives the async API behind the synchronous C ABI; the entity set is minimal (projects table: id, name, payload blob, version, timestamps). The graph payload is the same serialized form the ove-xml backend uses — one serialization truth, two containers.
- No callbacks/events (M10: synchronous commands only; the caller — oaktask/facade — owns progress reporting).
- Errors follow the project -MMCCCC scheme, module 10
(
-100001…); the M10 positive info codes (TOO_OLD/TOO_NEW/…) are kept verbatim. - Interchange is lossy. The otio backend's export/import mapping
preserves sequences/tracks/clips/gaps/transitions; effect chains,
keyframes, project bins/settings and exact rational timebases are
not carried (see the module docs in
backends/otio.rs).
Layout
src/
lib.rs crate doc + module map
error.rs error/info codes (M10 §2.1, -MMCCCC module 10)
handle.rs shared CHandle re-export (boxes live in oaknode's handle.rs)
uri.rs URI parsing/classification
session.rs StorageProject session (open/take/uri)
registry.rs backend registry (register/unregister/arbitrate)
backend.rs StorageBackend trait + LoadResult
backends/
ove_xml.rs built-in .ove XML backend (via bridge::node)
otio.rs built-in .otio/.fcpxml backend (via oakotio)
database.rs declared stub (later proxy)
bridge/
node.rs oaknode calls (project + serializer + sequence builder)
ffi.rs export layer (M10 §2.2/§2.3 verbatim)
tests/ contract tests incl. the pluggability proof
Build / test
A workspace member (not a default member); build and test explicitly:
cargo test -p oakstorage