dtourolle/jellytau
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases 38 Wiki Activity

feat(playback): let Rust decide what stream to play, and say so

Browse Source 
Playing a video meant asking the server to re-encode it, always. That
decision was made nowhere and written down nowhere, so whoever needed it
re-derived it downstream — the player worked out whether it had been handed
a playlist by looking for ".m3u8" in the URL, in two places. A viewer paid
for a transcode of a file their device could have played untouched, and the
app could not tell them which it was.

One negotiation now produces one self-describing StreamSelection — direct
play, remux or transcode; over a playlist, a plain HTTP file, or a local one
— and every renderer consumes that same answer.

Measured against the development server (Jellyfin 10.11.5), 400 items
sampled for codec mix and 40 put through a real PlaybackInfo negotiation
per profile:

  Linux / WebKitGTK (h264 only, 2ch)          3/40 —  7% direct play
  Android / ExoPlayer (hevc, ac3/eac3, 6ch)  34/40 — 85% direct play

The library is ~80% hevc, which is why the two diverge so hard. The payoff
is overwhelmingly Android, where 85% of plays were starting a transcode
nobody needed. Linux stays near 7% until libmpv decodes the picture — the
h264-only profile is a WebKitGTK constraint, not a JellyTau choice.

DR-219  StreamSelection: url + tagged Transport (hls/progressive/localFile)
        + PlaybackKind (directPlay/directStream/transcode) + the negotiated
        rendition + this source's ladder + a needs_transcoding flag derived
        in Rust so the rule is answered once. Both enums are serde-tagged
        so the frontend matches a discriminant, not a substring. The paths
        that never negotiate get the same shape from Rust rather than
        assembling one — media_local_selection for a downloaded file,
        LiveStreamInfo.transport for a live channel — so there is no second
        place where a transport is decided.

DR-220  The ceiling becomes two levels: a durable device default (Settings,
        persisted) and a per-playback override the in-player picker sets.
        The picker had called itself a "this film, this connection" control
        since it was written but wrote the process-wide default, so dropping
        one awkward film to 2 Mbps silently capped every video played
        afterwards for the rest of the process, with Settings still showing
        the old value. The override is cleared whenever playback moves to a
        new item, which stops it surviving into an autoplayed next episode.
        effective_streaming_quality() is the single resolution point.

DR-221  The quality picker is filled from what this media source can offer.
        Rust marks a rung exceeds_source when its ceiling is at or above the
        source's own bitrate — such a rung is another way to spell Original
        — and the frontend does not draw those. Original is never marked; a
        source whose bitrate the server does not report marks nothing, which
        keeps every rung offered.

DR-222  Direct play and direct stream are negotiated, with two client-side
        overrides on top because the server's answer is right about the file
        and wrong about what this app will do with it: undecodable audio
        (Jellyfin 10.11.5 honours a DirectPlayProfile's container and video
        codec but ignores its audio codec, so it offers direct play for an
        E-AC-3 track the webview renders in silence) and a viewer-pinned
        audio track the file does not default to. A direct stream is a remux
        and is deliberately not counted as transcoding.

DR-223  Dropped on measurement, not deferred. A master playlist from this
        server carries exactly one EXT-X-STREAM-INF: Jellyfin builds it from
        the single rendition the request asked for rather than publishing a
        ladder. So there is no adaptation for hls.js to be preserving and
        none mpv would lose — the claim that there was, in
        playback-backend-unification.md, does not hold. Recorded rather than
        deleted because it is a measurement: a server that does publish a
        ladder would change the answer.

DR-224  Every backend consumes the same selection. The queue item carries
        the transport, so player_seek_video picks its seek strategy from the
        backend's decision instead of the last stream_url.contains(".m3u8")
        in the codebase. Items queued by a path that never negotiated carry
        None and fall back to needs_transcoding, which is exact rather than
        a guess because every transcode this app requests is HLS (DR-140).

The frontend loader decision moves to streamTransport.ts so it can be
tested: the two cases that pin it are the ones that failed against the old
implementation — a progressive stream whose URL contains ".m3u8" must not
get an HLS loader, and an HLS stream whose URL contains none must.

Also verified the URL the direct-play branch builds actually serves playable
bytes: 206, video/mp4, valid ISO-BMFF, and a mid-file range works, so
seeking a direct play works.

The spec is folded into docs/architecture/{01,02,03} and deleted, per the
rule that docs/specs holds only work that has not shipped. DR-121 leaves
read-through-media-cache.md with a pointer; that spec keeps its capture half.

Not verified: real playback on a device. Direct play changes what actually
gets played, and neither fixtures nor curl prove the WebKitGTK and ExoPlayer
paths render it.
This commit is contained in:
dtourolle
2026-08-22 13:45:03 +02:00
parent 5fede123e7
commit 109700b949
45 changed files with 9744 additions and 6581 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/architecture/01-rust-backend.md
+144 -1
View File
@@ -673,8 +673,151 @@ device profile. Sending it there — not just on the transcode URL — is what m
the cap real: a stream the server decides to *direct play* is served at the
source file's own bitrate, and no URL parameter afterwards can reduce it.
#### Two levels of ceiling
**Location**: `src-tauri/src/repository/online.rs` (TRACES: UR-074, UR-079 | DR-225)
There are two, and they are not the same thing:
| | Set by | Lives until | Read via |
|---|---|---|---|
| **Device default** | Settings (`player_set_video_settings`) | Persisted; restored at startup | `streaming_quality()` |
| **Per-playback override** | The in-player picker (`player_set_stream_quality`) | The next item starts playing | `playback_quality_override()` |
`effective_streaming_quality()` resolves the pair — override first, else default —
and **is the only thing stream construction may read**. Every URL builder and the
`PlaybackInfo` negotiation go through it, for the reason the process-wide static
existed in the first place: if the negotiation and the URL builder disagree, the
cap leaks — the negotiation authorises a direct play the builder then never gets
to constrain, or the reverse.
> The override exists because a single global cannot express "this 4K remux needs
> a ceiling, that podcast does not". The picker had documented itself as a "this
> film, this connection" control since it was written, but was implemented by
> writing the *default* — so dropping one awkward film to 2 Mbps silently capped
> every video played afterwards for the rest of the process, with Settings still
> showing the old value. It is cleared on every `player_play_item` /
> `player_play_queue` / `player_play_tracks`, which is what stops it surviving
> into an autoplayed next episode where nobody would reopen the picker.
### Stream selection
**Location**: `src-tauri/src/repository/stream_selection.rs`,
`OnlineRepository::get_stream_selection` (TRACES: UR-070, UR-079 | DR-224, DR-226, DR-227)
**Rust decides *what stream*. The player decides *how to deliver it*.** That line
is the whole design. A backend with genuine adaptive selection (ExoPlayer over a
multi-variant playlist) is left to do it; Rust chooses what to request and never
paces bytes.
`get_stream_selection` returns one self-describing `StreamSelection` in place of
the bare URL `get_video_stream_url` used to hand out:
| Field | Carries |
|---|---|
| `url` | What to open |
| `transport` | `Hls` / `Progressive` / `LocalFile` — how to fetch it |
| `playback_kind` | `DirectPlay` / `DirectStream` / `Transcode` — what the server is doing to the source |
| `rendition` | The negotiated ceiling and codecs; `None` for a direct play, which *is* the source |
| `available` | The quality ladder as it applies to this media source (DR-226) |
| `needs_transcoding` | Derived from `playback_kind`, so the rule is answered once |
Both enums are serde-tagged (`{"type":"hls"}`) so the frontend matches a
discriminant rather than comparing text.
> **Why `transport` exists.** `VideoPlayer.svelte` chose its loader with
> `url.includes(".m3u8")`, in two places. Rust *built* that URL and knows exactly
> what it is; re-deriving it downstream by substring match is a domain fact
> reconstructed in the presentation layer — the same class of error as leaking
> item-type taxonomy, and one that fails silently in **both** directions: a
> progressive file served from a path containing the substring gets an HLS
> loader, and a playlist served from a path without it does not.
>
> The paths that never negotiate get the same shape from Rust rather than letting
> a caller assemble one — `media_local_selection` for a downloaded file,
> `LiveStreamInfo.transport` for a live channel — so there is no second place
> where a transport is decided.
#### The playback-kind decision
`decide_playback_kind` is a free function and pure, so every branch is testable
from `PlaybackInfo` fixtures without a server. Order matters — the two
client-side overrides come first, because each describes a case where the
server's answer is right about the *file* and wrong about what this app will do
with it:
1. **Undecodable audio → `Transcode`.** Jellyfin 10.11.5 honours a
DirectPlayProfile's container and video codec but *ignores its audio codec*,
so it offers direct play for an E-AC-3 track the webview renders in silence.
A silent direct play is worse than a transcode.
2. **A pinned audio track → `Transcode`.** Not a defect in the server's answer, a
different question: the file has one default track and the viewer asked for
another.
3. Otherwise `supports_direct_play` → `DirectPlay`, else `supports_direct_stream`
→ `DirectStream`, else `Transcode`.
A direct **stream** is a remux — codecs copied, container repackaged. It is cheap
and is deliberately *not* counted as transcoding; conflating the two would report
a free passthrough as a server-side re-encode.
> **What this is worth, measured.** Against the development server (Jellyfin
> 10.11.5), 400 items sampled for codec mix and 40 put through a real negotiation
> per profile:
>
> | Profile | Direct play |
> |---|---|
> | Linux / WebKitGTK (`h264` only, 2ch) | 3/40 — **7%** |
> | Android / ExoPlayer (`h264,hevc,vp8,vp9,av1,mpeg4` + `ac3,eac3`, 6ch) | 34/40 — **85%** |
>
> The library is ~80% hevc (`hevc+eac3` alone is a third of it), which is why the
> two diverge so hard. **The payoff is overwhelmingly Android**, where 85% of
> plays previously burned a transcode nobody needed. Linux stays near 7% until
> libmpv decodes the picture — the h264-only profile is a WebKitGTK constraint,
> not a JellyTau choice, and is what `linux-native-video-spike.md` exists to
> remove. A reviewer should not expect this code to fix Linux on its own.
#### The quality ladder per source
`quality_options_for_source(source_bitrate)` returns every rung, each marked with
`exceeds_source`: true when that rung's ceiling is at or above what the source
itself carries, so selecting it produces the same bytes as `Original`. The
frontend draws the list and drops the redundant rungs; it does not decide which
they are.
- `Original` is never marked — it *is* the source.
- An unreported source bitrate (some containers have none; the sampled library
has `avi` files with no bitrate at all) marks **nothing** redundant, keeping
every rung offered. That is the safe direction: the viewer keeps every choice.
#### No adaptive ladder to preserve
**TRACES: UR-079 | DR-228 (Won't Do)**
Mid-playback re-negotiation on throughput was scoped and dropped on measurement.
A master playlist from this server carries exactly **one** `EXT-X-STREAM-INF`:
Jellyfin builds it from the single rendition the request asked for rather than
publishing a ladder. So there is no adaptation for hls.js to be preserving and
none that mpv would lose — the claim that there was is recorded in
`playback-backend-unification.md` and does not hold. "Adapt mid-stream" collapses
into "pick well at open", which is what the two levels of ceiling and the
per-source ladder already are.
Kept here because it is a measurement, not an opinion: a server that *does*
publish a ladder would change the answer, and the re-negotiation path below is
the hook that work would build on.
#### Re-negotiation
One mechanism, not two. `player_seek_video`, `player_switch_audio_track` and
`player_set_stream_quality` all return a tagged `strategy` saying who reloads —
the backend handles a native backend itself and hands the webview a
`StreamSelection` for `reloadSource`. Note the wire wart: tauri-specta keeps
these response fields snake_case (`seek_offset`), while the `strategy` tag itself
is camelCase.
The frontend names a variant and nothing else; the labels the picker shows are
served over IPC by `player_get_streaming_qualities`.
served over IPC — from `available` on the selection, or
`player_get_streaming_qualities` for the Settings list.
## Background workers