Crate mediaschema 

Product-agnostic media-primitive schema.

The architectural hub is the hand-written domain layer; every backend — protobuf wire (under [buffa]), sqlx row mappers (under [sqlx]), MongoDB (under [mongodb]) — is a thin lossless conversion to/from the domain. Locked schema docs live under schema/*.md; they are the specification the implementation tracks.

Encode-side bridges (From<&Domain> for Backend) are infallible — the domain is the validated side. Decode-side bridges (TryFrom<Backend> for Domain) route through the same try_new + with_* builders application code uses, so every invariant is re-enforced at the wire/storage edge.

Aggregate clusters

• domain::Media / domain::MediaFile / domain::WatchedLocation — content-addressed media identity + per-copy locator + FS-event monitor.
• domain::Audio / domain::AudioTrack / domain::AudioSegment (+ Word child) — audio cluster.
• domain::Video / domain::VideoTrack / domain::Scene / domain::Keyframe — video cluster.
• domain::Subtitle / domain::SubtitleTrack / domain::SubtitleCue<Id, D> — subtitle cluster; cues are polymorphic over the per-format payload D (SrtData, VttData, AssData, LrcData).
• domain::Person / domain::Speaker — identity (per-track diarized voice with embedding-keyed voiceprint).

Feature flags

Three independent capability tiers (alloc / std, both additive), three medium-aggregate gates (video / audio / subtitle), plus optional backend features. Defaults are std + video + audio + subtitle.

flag	role	depends on
none	no-std + no-alloc; stack-only types only	—
alloc	no-std + alloc; adds heap-using domain types	—
std (default)	adds jiff-using aggregates + Uuid::now_v7	—
video (default)	compiles the Video / VideoTrack / Scene / Keyframe aggregate tree + its backends	a heap tier
audio (default)	compiles the Audio / AudioTrack / AudioSegment aggregate tree + its backends	a heap tier
subtitle (default)	compiles the Subtitle / SubtitleTrack / SubtitleCue aggregate tree + its backends	a heap tier
buffa	proto3 wire layer (under [buffa])	std or alloc
json	wire JSON via serde	std + buffa
arbitrary	Arbitrary derives for property tests	std + buffa
mongodb	bson backend (under [mongodb])	std + json
sqlx-postgres / sqlx-mysql / sqlx-sqlite	sql backends (under [sqlx])	std

The three medium gates are independent on/off flags: a consumer that only needs the video aggregate tree can opt out of the audio / subtitle trees (and all their backend bridges) with default-features = false + features = ["std", "video"]. The cross-cutting aggregates (Media, MediaFile, Person, Speaker, WatchedLocation, UserTag, SceneAnnotation) plus the Identified<Id, D> transport envelope are always available when a heap tier is on.

Regenerating wire code

The buffa-generated wire layer (src/generated/) is produced from the .proto files in proto/. Regenerate with cargo run -p xtask -- gen. Do not hand-edit the generated files.

Re-exports

pub use crate::domain::Identified;

Modules

domain

Hand-written domain layer — the architectural hub. App logic programs against mediaschema::domain::*; the buffa-generated wire types at the crate root (re-exported from generated::media::v1::*, available with feature = "std") are the serialization edge. Domain ⇄ wire conversions are added incrementally as each aggregate lands (with domain → wire → domain round-trip property tests). See schema/*.md for the locked specifications. Hand-written domain layer — the architectural hub for mediaschema.