Bug 771074 - Introduce pointer wrapping mechanism for hold references to pointers off-main-thread. r=bsmedberg

xpcom/glue/nsProxyRelease.h

@@ -9,6 +9,7 @@
 #include "nsIEventTarget.h"
 #include "nsCOMPtr.h"
 #include "nsAutoPtr.h"
+#include "nsThreadUtils.h"
 
 #ifdef XPCOM_GLUE_AVOID_NSPR
 #error NS_ProxyRelease implementation depends on NSPR.
@@ -61,4 +62,118 @@ NS_COM_GLUE nsresult
 NS_ProxyRelease
     (nsIEventTarget *target, nsISupports *doomed, bool alwaysProxy=false);
 
+/**
+ * Class to safely handle main-thread-only pointers off the main thread.
+ *
+ * Classes like XPCWrappedJS are main-thread-only, which means that it is
+ * forbidden to call methods on instances of these classes off the main thread.
+ * For various reasons (see bug 771074), this restriction recently began to
+ * apply to AddRef/Release as well.
+ *
+ * This presents a problem for consumers that wish to hold a callback alive
+ * on non-main-thread code. A common example of this is the proxy callback
+ * pattern, where non-main-thread code holds a strong-reference to the callback
+ * object, and dispatches new Runnables (also with a strong reference) to the
+ * main thread in order to execute the callback. This involves several AddRef
+ * and Release calls on the other thread, which is (now) verboten.
+ *
+ * The basic idea of this class is to introduce a layer of indirection.
+ * nsMainThreadPtrHolder is a threadsafe reference-counted class that internally
+ * maintains one strong reference to the main-thread-only object. It must be
+ * instantiated on the main thread (so that the AddRef of the underlying object
+ * happens on the main thread), but consumers may subsequently pass references
+ * to the holder anywhere they please. These references are meant to be opaque
+ * when accessed off-main-thread (assertions enforce this).
+ *
+ * The semantics of nsRefPtr<nsMainThreadPtrHolder<T> > would be cumbersome, so
+ * we also introduce nsMainThreadPtrHandle<T>, which is conceptually identical
+ * to the above (though it includes various convenience methods). The basic
+ * pattern is as follows.
+ *
+ * // On the main thread:
+ * nsCOMPtr<nsIFooCallback> callback = ...;
+ * nsMainThreadPtrHandle<nsIFooCallback> callbackHandle =
+ *   new nsMainThreadPtrHolder<nsIFooCallback>(callback);
+ * // Pass callbackHandle to structs/classes that might be accessed on other
+ * // threads.
+ *
+ * All structs and classes that might be accessed on other threads should store
+ * an nsMainThreadPtrHandle<T> rather than an nsCOMPtr<T>.
+ */
+template<class T>
+class nsMainThreadPtrHolder
+{
+public:
+  // We can only acquire a pointer on the main thread.
+  nsMainThreadPtrHolder(T* ptr) : mRawPtr(NULL) {
+    // We can only AddRef our pointer on the main thread, which means that the
+    // holder must be constructed on the main thread.
+    MOZ_ASSERT(NS_IsMainThread());
+    NS_IF_ADDREF(mRawPtr = ptr);
+  }
+
+  // We can be released on any thread.
+  ~nsMainThreadPtrHolder() {
+    if (NS_IsMainThread()) {
+      NS_IF_RELEASE(mRawPtr);
+    } else {
+      nsCOMPtr<nsIThread> mainThread = do_GetMainThread();
+      if (!mainThread) {
+        NS_WARNING("Couldn't get main thread! Leaking pointer.");
+        return;
+      }
+      NS_ProxyRelease(mainThread, mRawPtr);
+    }
+  }
+
+  T* get() {
+    // Nobody should be touching the raw pointer off-main-thread.
+    if (NS_UNLIKELY(!NS_IsMainThread()))
+      MOZ_CRASH();
+    return mRawPtr;
+  }
+
+  NS_IMETHOD_(nsrefcnt) Release();
+  NS_IMETHOD_(nsrefcnt) AddRef();
+
+private:
+  // This class is threadsafe and reference-counted.
+  nsAutoRefCnt mRefCnt;
+
+  // Our wrapped pointer.
+  T* mRawPtr;
+
+  // Copy constructor and operator= not implemented. Once constructed, the
+  // holder is immutable.
+  T& operator=(nsMainThreadPtrHolder& other);
+  nsMainThreadPtrHolder(const nsMainThreadPtrHolder& other);
+};
+
+template<class T>
+NS_IMPL_THREADSAFE_ADDREF(nsMainThreadPtrHolder<T>)
+template<class T>
+NS_IMPL_THREADSAFE_RELEASE(nsMainThreadPtrHolder<T>)
+
+template<class T>
+class nsMainThreadPtrHandle
+{
+  nsRefPtr<nsMainThreadPtrHolder<T> > mPtr;
+
+  public:
+  nsMainThreadPtrHandle(nsMainThreadPtrHolder<T> *aHolder) : mPtr(aHolder) {}
+  nsMainThreadPtrHandle(const nsMainThreadPtrHandle& aOther) : mPtr(aOther.mPtr) {}
+  nsMainThreadPtrHandle& operator=(const nsMainThreadPtrHandle& aOther) {
+    mPtr = aOther.mPtr;
+  }
+
+  operator nsMainThreadPtrHolder<T>*() { return mPtr.get(); }
+
+  // These all call through to nsMainThreadPtrHolder, and thus implicitly
+  // assert that we're on the main thread. Off-main-thread consumers must treat
+  // these handles as opaque.
+  T* get() { return mPtr.get()->get(); }
+  operator T*() { return get(); }
+  T* operator->() { return get(); }
+};
+
 #endif