Prevent realtime scheduling for ad-hoc acceptors

I noticed that there were some realtime adhoc-acceptors running on my
system. That should of course not happen, since these only exist to
catch some sporadic (and likely as a result of a badly behaving plugin)
mutual recursion on the audio thread.

CHANGELOG.md

@@ -16,6 +16,10 @@ Versioning](https://semver.org/spec/v2.0.0.html).
   and
   [!1120](https://gitlab.freedesktop.org/pipewire/pipewire/-/merge_requests/1120)
   have been merged.
+- Prevented yabridge's ad-hoc socket acceptors from inheriting realtime
+  scheduling when spawned from audio threads. In practice this should not make
+  any difference as these threads are sleeping all the time except for under
+  very specific circumstances.
 
 ### Fixed
 

src/common/communication/common.h

@@ -780,6 +780,9 @@ class AdHocSocketHandler {
         Thread secondary_requests_handler([&]() {
             pthread_setname_np(pthread_self(), "adhoc-acceptor");
 
+            // Any secondary threads should not be realtime
+            set_realtime_priority(false);
+
             secondary_context.run();
         });
 

src/common/utils.h

@@ -102,6 +102,12 @@ std::optional<int> get_realtime_priority() noexcept;
  *
  * @return Whether the operation was successful or not. This will fail if the
  *   user does not have the privileges to set realtime priorities.
+ *
+ * TODO: At some point, consider using `SCHED_RESET_ON_FORK` instead of manually
+ *       disabling this when we don't want realtime scheduling to propagate.
+ *       That would require a bit of careful analysis because we do want it to
+ *       propagate to a Windows plugin's audio threads, and I don't think
+ *       there's a way to go back once you've set `SCHED_RESET_ON_FORK`.
  */
 bool set_realtime_priority(bool sched_fifo, int priority = 5) noexcept;