Bump sendspin-cpp pin to v0.8.0

c_src/sendspin_player/CMakeLists.txt

@@ -21,7 +21,7 @@ set(CMAKE_CXX_FLAGS_RELEASE "-O3 -DNDEBUG" CACHE STRING "" FORCE)
 # build dirs. This also means `-DSENDSPIN_CPP_REF=...` cannot override;
 # for scratch experiments against another ref use
 # `-DFETCHCONTENT_SOURCE_DIR_SENDSPIN-CPP=<local-checkout>` instead.
-set(SENDSPIN_CPP_REF "30514d5102c269a0c7fa6a13932d6bf7f2ae1abc" # v0.7.2
+set(SENDSPIN_CPP_REF "9331ace6428979982c934384702d289958ca125e" # v0.8.0
     CACHE STRING "sendspin-cpp commit SHA to pin" FORCE)
 
 # Trim sendspin-cpp to just the player role. Other roles ship dead code on a

c_src/sendspin_player/README.md

@@ -69,8 +69,8 @@ Differences:
 ## sendspin-cpp pin
 
 Pinned via `FetchContent` to a **commit SHA** in `CMakeLists.txt`
-(`SENDSPIN_CPP_REF` cache var). Currently `30514d51...`, upstream
-`v0.7.2`. We pin to a SHA rather
+(`SENDSPIN_CPP_REF` cache var). Currently `9331ace6...`, upstream
+`v0.8.0`. We pin to a SHA rather
 than a tag because git tags are mutable server-side; a retagged upstream
 would otherwise silently flow into firmware. The friendly tag name (or
 nearest-release note) lives in a comment beside the SHA for traceability.

c_src/sendspin_player/src/main.cpp

@@ -653,7 +653,7 @@ struct ClientListener : SendspinClientListener {
 
 // sendspin-cpp binds its WebSocket listener lazily — on the first
 // client.loop() tick after the network provider reports ready — so
-// start_server() returning does NOT mean the port accepts connections
+// start() returning does NOT mean the port accepts connections
 // yet. The Elixir side must not advertise the player over mDNS before
 // the listener is up: Music Assistant's discovery connect is one-shot
 // (aiosendspin `retry_initial_connection=False`), so a connection
@@ -792,9 +792,13 @@ int main(int argc, char* argv[]) {
         emit_json(os.str());
     }
 
-    if (!client.start_server()) {
-        emit_json("{\"event\":\"error\",\"kind\":\"start_server\","
-                  "\"msg\":\"failed to bind WebSocket listener\"}");
+    // A false return means a role failed to start (thread spawn, ring
+    // buffer allocation); it does NOT mean the listener failed to bind,
+    // which cannot be known yet — start() only arms the server, and the
+    // bind happens on a later loop() tick (see the hook comment above).
+    if (!client.start()) {
+        emit_json("{\"event\":\"error\",\"kind\":\"start\","
+                  "\"msg\":\"failed to start client roles\"}");
         return 1;
     }
 
@@ -874,7 +878,16 @@ int main(int argc, char* argv[]) {
         std::this_thread::sleep_for(std::chrono::milliseconds(10));
     }
 
-    client.disconnect(SendspinGoodbyeReason::SHUTDOWN);
+    // stop() rather than disconnect(): it sends the shutdown goodbye AND
+    // joins the role threads before returning. That ordering is load-
+    // bearing here — `client` is declared before `audio_sink` and the
+    // listeners, so reverse destruction would otherwise tear those down
+    // while ~SendspinClient() has yet to join the threads that call into
+    // them. Every PlayerListener callback dereferences `sink` (including
+    // on_audio_write on the role thread's hot path), so a late callback
+    // would touch a destroyed stack object. Upstream states the contract
+    // directly: listeners must outlive the client.
+    client.stop();
     emit_json("{\"event\":\"shutdown\"}");
     return 0;
 }

test/sendspin_player_contract_test.exs

@@ -7,8 +7,9 @@ defmodule SendspinPlayerContractTest do
 
   Spawns the host build of the binary, drives stdin/stdout, and asserts
   the documented event/command shapes. Does not require audio hardware —
-  the binary's `start_server()` binds a non-privileged WebSocket port and
-  does not touch ALSA until a stream begins.
+  the binary listens on a non-privileged WebSocket port (`start()` arms
+  the server; the bind lands on a later `loop()` tick) and does not touch
+  ALSA until a stream begins.
   """
 
   use ExUnit.Case, async: false