5839b36192
This was a huge effort by photovoltex@gmail.com with help from the community. Over 140 commits were squashed. Below, their commit messages are kept unchanged. --- * dealer wrapper for ease of use * improve sending protobuf requests * replace connect config with connect_state config * start integrating dealer into spirc * payload handling, gzip support * put connect state consistent * formatting * request payload handling, gzip support * expose dealer::protocol, move request in own file * integrate handle of connect-state commands * spirc: remove ident field * transfer playing state better * spirc: remove remote_update stream * spirc: replace command sender with connect state update * spirc: remove device state and remaining unused methods * spirc: remove mercury sender * add repeat track state * ConnectState: add methods to replace state in spirc * spirc: move context into connect_state, update load and next * spirc: remove state, adjust remaining methods * spirc: handle more dealer request commands * revert rustfmt.toml * spirc: impl shuffle - impl shuffle again - extracted fill up of next tracks in own method - moved queue revision update into next track fill up - removed unused method `set_playing_track_index` - added option to specify index when resetting the playback context - reshuffle after repeat context * spirc: handle device became inactive * dealer: adjust payload handling * spirc: better set volume handling * dealer: box PlayCommand (clippy warning) * dealer: always respect queued tracks * spirc: update duration of track * ConnectState: update more restrictions * cleanup * spirc: handle queue requests * spirc: skip next with track * proto: exclude spirc.proto - move "deserialize_with" functions into own file - replace TrackRef with ProvidedTrack * spirc: stabilize transfer/context handling * core: cleanup some remains * connect: improvements to code structure and performance - use VecDeque for next and prev tracks * connect: delayed volume update * connect: move context resolve into own function * connect: load context asynchronous * connect: handle reconnect - might currently steal the active devices playback * connect: some fixes and adjustments - fix wrong offset when transferring playback - fix missing displayed context in web-player - remove access_token from log - send correct state reason when updating volume - queue track correctly - fix wrong assumption for skip_to * connect: replace error case with option * connect: use own context state * connect: more stabilising - handle SkipTo having no Index - handle no transferred restrictions - handle no transferred index - update state before shutdown, for smoother reacquiring * connect: working autoplay * connect: handle repeat context/track * connect: some quick fixes - found self-named uid in collection after reconnecting * connect: handle add_to_queue via set_queue * fix clippy warnings * fix check errors, fix/update example * fix 1.75 specific error * connect: position update improvements * connect: handle unavailable * connect: fix incorrect status handling for desktop and mobile * core: fix dealer reconnect - actually acquire new token - use login5 token retrieval * connect: split state into multiple files * connect: encapsulate provider logic * connect: remove public access to next and prev tracks * connect: remove public access to player * connect: move state only commands into own file * connect: improve logging * connect: handle transferred queue again * connect: fix all-features specific error * connect: extract transfer handling into own file * connect: remove old context model * connect: handle more transfer cases correctly * connect: do auth_token pre-acquiring earlier * connect: handle play with skip_to by uid * connect: simplified cluster update log * core/connect: add remaining set value commands * connect: position update workaround/fix * connect: some queue cleanups * connect: add uid to queue * connect: duration as volume delay const * connect: some adjustments and todo cleanups - send volume update before general update - simplify queue revision to use the track uri - argument why copying the prev/next tracks is fine * connect: handle shuffle from set_options * connect: handle context update * connect: move other structs into model.rs * connect: reduce SpircCommand visibility * connect: fix visibility of model * connect: fix: shuffle on startup isn't applied * connect: prevent loading a context with no tracks * connect: use the first page of a context * connect: improve context resolving - support multiple pages - support page_url of context - handle single track * connect: prevent integer underflow * connect: rename method for better clarity * connect: handle mutate and update messages * connect: fix 1.75 problems * connect: fill, instead of replace next page * connect: reduce context update to single method * connect: remove unused SpircError, handle local files * connect: reduce nesting, adjust initial transfer handling * connect: don't update volume initially * core: disable trace logging of handled mercury responses * core/connect: prevent takeover from other clients, handle session-update * connect: add queue-uid for set_queue command * connect: adjust fields for PlayCommand * connect: preserve context position after update_context * connect: unify metadata modification - only handle `is_queued` `true` items for queue * connect: polish request command handling - reply to all request endpoints - adjust some naming - add some docs * connect: add uid to tracks without * connect: simpler update of current index * core/connect: update log msg, fix wrong behavior - handle became inactive separately - remove duplicate stop - adjust docs for websocket request * core: add option to request without metrics and salt * core/context: adjust context requests and update - search should now return the expected context - removed workaround for single track playback - move local playback check into update_context - check track uri for invalid characters - early return with `?` * connect: handle possible search context uri * connect: remove logout support - handle logout command - disable support for logout - add todos for logout * connect: adjust detailed tracks/context handling - always allow next - handle no prev track available - separate active and fill up context * connect: adjust context resolve handling, again * connect: add autoplay metadata to tracks - transfer into autoplay again * core/connect: cleanup session after spirc stops * update CHANGELOG.md * playback: fix clippy warnings * connect: adjust metadata - unify naming - move more metadata infos into metadata.rs * connect: add delimiter between context and autoplay playback * connect: stop and resume correctly * connect: adjust context resolving - improved certain logging parts - preload autoplay when autoplay attribute mutates - fix transfer context uri - fix typo - handle empty strings for resolve uri - fix unexpected stop of playback * connect: ignore failure during stop * connect: revert resolve_uri changes * connect: correct context reset * connect: reduce boiler code * connect: fix some incorrect states - uid getting replaced by empty value - shuffle/repeat clearing autoplay context - fill_up updating and using incorrect index * core: adjust incorrect separator * connect: move `add_to_queue` and `mark_unavailable` into tracks.rs * connect: refactor - directly modify PutStateRequest - replace `next_tracks`, `prev_tracks`, `player` and `device` with `request` - provide helper methods for the removed fields * connect: adjust handling of context metadata/restrictions * connect: fix incorrect context states * connect: become inactive when no cluster is reported * update CHANGELOG.md * core/playback: preemptively fix clippy warnings * connect: minor adjustment to session changed * connect: change return type changing active context * connect: handle unavailable contexts * connect: fix previous restrictions blocking load with shuffle * connect: update comments and logging * core/connect: reduce some more duplicate code * more docs around the dealer
3.6 KiB
Dealer
When talking about the dealer, we are speaking about a websocket that represents the player as spotify-connect device. The dealer is primarily used to receive updates and not to update the state.
Messages and Requests
There are two types of messages that are received via the dealer, Messages and Requests. Messages are fire-and-forget and don't need a responses, while request expect a reply if the request was processed successfully or failed.
Because we publish our device with support for gzip, the message payload might be BASE64 encoded and gzip compressed. If that is the case, the related headers send an entry for "Transfer-Encoding" with the value of "gzip".
Messages
Most messages librespot handles send bytes that can be easily converted into their respective
protobuf definition. Some outliers send json that can be usually mapped to an existing protobuf
definition. We use protobuf-json-mapping to a similar protobuf definition
Note: The json sometimes doesn't map exactly and can provide more fields than the protobuf definition expects. For messages, we usually ignore unknown fields.
There are two types of messages, "informational" and "fire and forget commands".
Informational:
Informational messages send any changes done by the current user or of a client where the current user is logged in. These messages contain for example changes to a own playlist, additions to the liked songs or any update that a client sends.
Fire and Forget commands:
These are messages that send information that are requests to the current player. These are only send to the active player. Volume update requests and the logout request are send as fire-forget-commands.
Requests
The request payload is sent as json. There are almost usable protobuf definitions (see
files named like es_<command in snakecase>(_request).proto) for the commands, but they don't
align up with the expected values and are missing some major information we need for handling some
commands. Because of that we have our own model for the specific commands, see
core/src/dealer/protocol/request.rs.
All request modify the player-state.
Details
This sections is for details and special hiccups in regards to handling that isn't completely intuitive.
UIDs
A spotify item is identifiable by their uri. The ContextTrack and ProvidedTrack both have a uid
field. When we receive a context via the context-resolver it can return items (ContextTrack) that
may have their respective uid set. Some context like the collection and albums don't provide this
information.
When a uid is missing, resorting the next tracks in an official client gets confused and sends
incorrect data via the set_queue request. To prevent this behavior we generate a uid for each
track that doesn't have an uid. Queue items become a "queue-uid" which is just a q with an
incrementing number.
Metadata
For some client's (especially mobile) the metadata of a track is very important to display the context correct. For example the "autoplay" metadata is relevant to display the correct context info.
Metadata can also be used to store data like the iteration when repeating a context.
Repeat
The context repeating implementation is partly mimicked from the official client. The official client allows skipping into negative iterations, this is currently not supported.
Repeating is realized by filling the next tracks with multiple contexts separated by delimiters. By that we only have to handle the delimiter when skipping to the next and previous track.