5 files changed, 565 insertions, 4 deletions

diff --git a/linden/doc/contributions.txt b/linden/doc/contributions.txt
index c8dd773..0ca56be 100644
--- a/linden/doc/contributions.txt
+++ b/linden/doc/contributions.txt
@@ -92,6 +92,7 @@ Aleric Inglewood
         IMP-670
         IMP-688
         IMP-692
+        IMP-701
         IMP-712
 Alissa Sabre
         VWR-81
diff --git a/linden/indra/llcommon/CMakeLists.txt b/linden/indra/llcommon/CMakeLists.txt
index a05ee6f..ee80ec3 100644
--- a/linden/indra/llcommon/CMakeLists.txt
+++ b/linden/indra/llcommon/CMakeLists.txt
@@ -76,6 +76,7 @@ set(llcommon_HEADER_FILES
     CMakeLists.txt
 
     aiaprpool.h
+    aithreadsafe.h
     bitpack.h
     ctype_workaround.h
     doublelinkedlist.h
diff --git a/linden/indra/llcommon/aiaprpool.h b/linden/indra/llcommon/aiaprpool.h
index f6d7ffa..72e9ddb 100644
--- a/linden/indra/llcommon/aiaprpool.h
+++ b/linden/indra/llcommon/aiaprpool.h
@@ -157,17 +157,19 @@ public:
 /**
  * @brief Root memory pool (allocates memory from the operating system).
  *
- * This class should only be used by AIThreadLocalData (and LLMutexRootPool
- * when APR_HAS_THREADS isn't defined).
+ * This class should only be used by AIThreadLocalData and AIThreadSafeSimpleDCRootPool_pbase
+ * (and LLMutexRootPool when APR_HAS_THREADS isn't defined).
  */
 class LL_COMMON_API AIAPRRootPool : public AIAPRInitialization, public AIAPRPool
 {
 private:
         friend class AIThreadLocalData;
+        friend class AIThreadSafeSimpleDCRootPool_pbase;
 #if !APR_HAS_THREADS
         friend class LLMutexRootPool;
 #endif
-        //! Construct a root memory pool. Should only be used by AIThreadLocalData.
+        //! Construct a root memory pool.
+        //  Should only be used by AIThreadLocalData and AIThreadSafeSimpleDCRootPool_pbase.
         AIAPRRootPool(void);
         ~AIAPRRootPool();
 
diff --git a/linden/indra/llcommon/aithreadsafe.h b/linden/indra/llcommon/aithreadsafe.h
new file mode 100644
index 0000000..e1b93ba
--- /dev/null
+++ b/linden/indra/llcommon/aithreadsafe.h
@@ -0,0 +1,463 @@
+/**
+ * @file aithreadsafe.h
+ * @brief Implementation of AIThreadSafe, AIReadAccessConst, AIReadAccess and AIWriteAccess.
+ *
+ * CHANGELOG
+ *   and additional copyright holders.
+ *
+ *   31/03/2010
+ *   Initial version, written by Aleric Inglewood @ SL
+ */
+
+#ifndef AITHREADSAFE_H
+#define AITHREADSAFE_H
+
+#include <new>
+
+#include "llthread.h"
+#include "llerror.h"
+
+template<typename T> struct AIReadAccessConst;
+template<typename T> struct AIReadAccess;
+template<typename T> struct AIWriteAccess;
+template<typename T> struct AIAccess;
+
+template<typename T>
+class AIThreadSafeBits
+{
+private:
+        // AIThreadSafe is a wrapper around an instance of T.
+        // Because T might not have a default constructor, it is constructed
+        // 'in place', with placement new, in the memory reserved here.
+        //
+        // Make sure that the memory that T will be placed in is properly
+        // aligned by using an array of long's.
+        long mMemory[(sizeof(T) + sizeof(long) - 1) / sizeof(long)];
+
+public:
+        // The wrapped objects are constructed in-place with placement new *outside*
+        // of this object (by AITHREADSAFE macro(s) or derived classes).
+        // However, we are responsible for the destruction of the wrapped object.
+        ~AIThreadSafeBits() { ptr()->~T(); }
+
+        // Only for use by AITHREADSAFE, see below.
+        void* memory() const { return const_cast<long*>(&mMemory[0]); }
+
+protected:
+        // Accessors.
+        T const* ptr() const { return reinterpret_cast<T const*>(mMemory); }
+        T* ptr() { return reinterpret_cast<T*>(mMemory); }
+};
+
+/**
+ * @brief A wrapper class for objects that need to be accessed by more than one thread, allowing concurrent readers.
+ *
+ * Use AITHREADSAFE to define instances of any type, and use AIReadAccessConst,
+ * AIReadAccess and AIWriteAccess to get access to the instance.
+ *
+ * For example,
+ *
+ * <code>
+ * class Foo { public: Foo(int, int); };
+ *
+ * AITHREADSAFE(Foo, foo, (2, 3));
+ *
+ * AIReadAccess<Foo> foo_r(foo);
+ * // Use foo_r-> for read access.
+ *
+ * AIWriteAccess<Foo> foo_w(foo);
+ * // Use foo_w-> for write access.
+ * </code>
+ *
+ * If <code>foo</code> is constant, you have to use <code>AIReadAccessConst<Foo></code>.
+ *
+ * It is possible to pass access objects to a function that
+ * downgrades the access, for example:
+ *
+ * <code>
+ * void readfunc(AIReadAccess const& access);
+ *
+ * AIWriteAccess<Foo> foo_w(foo);
+ * readfunc(foo_w);                                     // readfunc will perform read access to foo_w.
+ * </code>
+ *
+ * If <code>AIReadAccess</code> is non-const, you can upgrade the access by creating
+ * an <code>AIWriteAccess</code> object from it. For example:
+ *
+ * <code>
+ * AIWriteAccess<Foo> foo_w(foo_r);
+ * </code>
+ *
+ * This API is Robust(tm). If you try anything that could result in problems,
+ * it simply won't compile. The only mistake you can still easily make is
+ * to obtain write access to an object when it is not needed, or to unlock
+ * an object in between accesses while the state of the object should be
+ * preserved. For example:
+ *
+ * <code>
+ * // This resets foo to point to the first file and then returns that.
+ * std::string filename = AIWriteAccess<Foo>(foo)->get_first_filename();
+ *
+ * // WRONG! The state between calling get_first_filename and get_next_filename should be preserved!
+ *
+ * AIWriteAccess<Foo> foo_w(foo);       // Wrong. The code below only needs read-access.
+ * while (!filename.empty())
+ * {
+ *     something(filename);
+ *     filename = foo_w->next_filename();
+ * }
+ * </code>
+ *
+ * Correct would be
+ *
+ * <code>
+ * AIReadAccess<Foo> foo_r(foo);
+ * std::string filename = AIWriteAccess<Foo>(foo_r)->get_first_filename();
+ * while (!filename.empty())
+ * {
+ *     something(filename);
+ *     filename = foo_r->next_filename();
+ * }
+ * </code>
+ *
+ */
+template<typename T>
+class AIThreadSafe : public AIThreadSafeBits<T>
+{
+protected:
+        // Only these may access the object (through ptr()).
+        friend struct AIReadAccessConst<T>;
+        friend struct AIReadAccess<T>;
+        friend struct AIWriteAccess<T>;
+
+        // Locking control.
+        AIRWLock mRWLock;
+
+        // For use by AIThreadSafeDC
+        AIThreadSafe(void) { }
+        AIThreadSafe(AIAPRPool& parent) : mRWLock(parent) { }
+
+public:
+        // Only for use by AITHREADSAFE, see below.
+        AIThreadSafe(T* object) { llassert(object == AIThreadSafeBits<T>::ptr()); }
+};
+
+/**
+ * @brief Instantiate an static, global or local object of a given type wrapped in AIThreadSafe, using an arbitrary constructor.
+ *
+ * For example, instead of doing
+ *
+ * <code>
+ * Foo foo(x, y);
+ * static Bar bar;
+ * </code>
+ *
+ * One can instantiate a thread-safe instance with
+ *
+ * <code>
+ * AITHREADSAFE(Foo, foo, (x, y));
+ * static AITHREADSAFE(Bar, bar, );
+ * </code>
+ *
+ * Note: This macro does not allow to allocate such object on the heap.
+ *       If that is needed, have a look at AIThreadSafeDC.
+ */
+#define AITHREADSAFE(type, var, paramlist) AIThreadSafe<type> var(new (var.memory()) type paramlist)
+
+/**
+ * @brief A wrapper class for objects that need to be accessed by more than one thread.
+ *
+ * This class is the same as an AIThreadSafe wrapper, except that it can only
+ * be used for default constructed objects.
+ *
+ * For example, instead of
+ *
+ * <code>
+ * Foo foo;
+ * </code>
+ *
+ * One would use
+ *
+ * <code>
+ * AIThreadSafeDC<Foo> foo;
+ * </code>
+ *
+ * The advantage over AITHREADSAFE is that this object can be allocated with
+ * new on the heap. For example:
+ *
+ * <code>
+ * AIThreadSafeDC<Foo>* ptr = new AIThreadSafeDC<Foo>;
+ * </code>
+ *
+ * which is not possible with AITHREADSAFE.
+ */
+template<typename T>
+class AIThreadSafeDC : public AIThreadSafe<T>
+{
+public:
+        // Construct a wrapper around a default constructed object.
+        AIThreadSafeDC(void) { new (AIThreadSafe<T>::ptr()) T; }
+};
+
+/**
+ * @brief Read lock object and provide read access.
+ */
+template<typename T>
+struct AIReadAccessConst
+{
+        //! Internal enum for the lock-type of the AI*Access object.
+        enum state_type
+        {
+                readlocked,                     //!< A AIReadAccessConst or AIReadAccess.
+                read2writelocked,       //!< A AIWriteAccess constructed from a AIReadAccess.
+                writelocked,            //!< A AIWriteAccess constructed from a AIThreadSafe.
+                write2writelocked       //!< A AIWriteAccess constructed from (the AIReadAccess base class of) a AIWriteAccess.
+        };
+
+        //! Construct a AIReadAccessConst from a constant AIThreadSafe.
+        AIReadAccessConst(AIThreadSafe<T> const& wrapper)
+                : mWrapper(const_cast<AIThreadSafe<T>&>(wrapper)),
+                  mState(readlocked)
+                {
+                        mWrapper.mRWLock.rdlock();
+                }
+
+        //! Destruct the AI*Access object.
+        // These should never be dynamically allocated, so there is no need to make this virtual.
+        ~AIReadAccessConst()
+                {
+                        if (mState == readlocked)
+                                mWrapper.mRWLock.rdunlock();
+                        else if (mState == writelocked)
+                                mWrapper.mRWLock.wrunlock();
+                        else if (mState == read2writelocked)
+                                mWrapper.mRWLock.wr2rdlock();
+                }
+
+        //! Access the underlaying object for read access.
+        T const* operator->() const { return mWrapper.ptr(); }
+
+        //! Access the underlaying object for read access.
+        T const& operator*() const { return *mWrapper.ptr(); }
+
+protected:
+        //! Constructor used by AIReadAccess.
+        AIReadAccessConst(AIThreadSafe<T>& wrapper, state_type state)
+                : mWrapper(wrapper), mState(state) { }
+
+        AIThreadSafe<T>& mWrapper;      //!< Reference to the object that we provide access to.
+        state_type const mState;        //!< The lock state that mWrapper is in.
+
+private:
+        // Disallow copy constructing directly.
+        AIReadAccessConst(AIReadAccessConst const&);
+};
+
+/**
+ * @brief Read lock object and provide read access, with possible promotion to write access.
+ */
+template<typename T>
+struct AIReadAccess : public AIReadAccessConst<T>
+{
+        typedef typename AIReadAccessConst<T>::state_type state_type;
+        using AIReadAccessConst<T>::readlocked;
+
+        //! Construct a AIReadAccess from a non-constant AIThreadSafe.
+        AIReadAccess(AIThreadSafe<T>& wrapper) : AIReadAccessConst<T>(wrapper, readlocked) { this->mWrapper.mRWLock.rdlock(); }
+
+protected:
+        //! Constructor used by AIWriteAccess.
+        AIReadAccess(AIThreadSafe<T>& wrapper, state_type state) : AIReadAccessConst<T>(wrapper, state) { }
+
+        friend class AIWriteAccess<T>;
+};
+
+/**
+ * @brief Write lock object and provide read/write access.
+ */
+template<typename T>
+struct AIWriteAccess : public AIReadAccess<T>
+{
+        using AIReadAccessConst<T>::readlocked;
+        using AIReadAccessConst<T>::read2writelocked;
+        using AIReadAccessConst<T>::writelocked;
+        using AIReadAccessConst<T>::write2writelocked;
+
+        //! Construct a AIWriteAccess from a non-constant AIThreadSafe.
+        AIWriteAccess(AIThreadSafe<T>& wrapper) : AIReadAccess<T>(wrapper, writelocked) { this->mWrapper.mRWLock.wrlock();}
+
+        //! Promote read access to write access.
+        explicit AIWriteAccess(AIReadAccess<T>& access)
+                : AIReadAccess<T>(access.mWrapper, (access.mState == readlocked) ? read2writelocked : write2writelocked)
+                {
+                        if (this->mState == read2writelocked)
+                        {
+                                this->mWrapper.mRWLock.rd2wrlock();
+                        }
+                }
+
+        //! Access the underlaying object for (read and) write access.
+        T* operator->() const { return this->mWrapper.ptr(); }
+
+        //! Access the underlaying object for (read and) write access.
+        T& operator*() const { return *this->mWrapper.ptr(); }
+};
+
+/**
+ * @brief A wrapper class for objects that need to be accessed by more than one thread.
+ *
+ * Use AITHREADSAFESIMPLE to define instances of any type, and use AIAccess
+ * to get access to the instance.
+ *
+ * For example,
+ *
+ * <code>
+ * class Foo { public: Foo(int, int); };
+ *
+ * AITHREADSAFESIMPLE(Foo, foo, (2, 3));
+ *
+ * AIAccess<Foo> foo_w(foo);
+ * // Use foo_w-> for read and write access.
+ *
+ * See also AIThreadSafe
+ */
+template<typename T>
+class AIThreadSafeSimple : public AIThreadSafeBits<T>
+{
+protected:
+        // Only this one may access the object (through ptr()).
+        friend struct AIAccess<T>;
+
+        // Locking control.
+        LLMutex mMutex;
+
+        // For use by AIThreadSafeSimpleDC
+        AIThreadSafeSimple(void) { }
+        AIThreadSafeSimple(AIAPRPool& parent) : mMutex(parent) { }
+
+public:
+        // Only for use by AITHREADSAFESIMPLE, see below.
+        AIThreadSafeSimple(T* object) { llassert(object == AIThreadSafeBits<T>::ptr()); }
+};
+
+/**
+ * @brief Instantiate an static, global or local object of a given type wrapped in AIThreadSafeSimple, using an arbitrary constructor.
+ *
+ * For example, instead of doing
+ *
+ * <code>
+ * Foo foo(x, y);
+ * static Bar bar;
+ * </code>
+ *
+ * One can instantiate a thread-safe instance with
+ *
+ * <code>
+ * AITHREADSAFESIMPLE(Foo, foo, (x, y));
+ * static AITHREADSAFESIMPLE(Bar, bar, );
+ * </code>
+ *
+ * Note: This macro does not allow to allocate such object on the heap.
+ *       If that is needed, have a look at AIThreadSafeSimpleDC.
+ */
+#define AITHREADSAFESIMPLE(type, var, paramlist) AIThreadSafeSimple<type> var(new (var.memory()) type paramlist)
+
+/**
+ * @brief A wrapper class for objects that need to be accessed by more than one thread.
+ *
+ * This class is the same as an AIThreadSafeSimple wrapper, except that it can only
+ * be used for default constructed objects.
+ *
+ * For example, instead of
+ *
+ * <code>
+ * Foo foo;
+ * </code>
+ *
+ * One would use
+ *
+ * <code>
+ * AIThreadSafeSimpleDC<Foo> foo;
+ * </code>
+ *
+ * The advantage over AITHREADSAFESIMPLE is that this object can be allocated with
+ * new on the heap. For example:
+ *
+ * <code>
+ * AIThreadSafeSimpleDC<Foo>* ptr = new AIThreadSafeSimpleDC<Foo>;
+ * </code>
+ *
+ * which is not possible with AITHREADSAFESIMPLE.
+ */
+template<typename T>
+class AIThreadSafeSimpleDC : public AIThreadSafeSimple<T>
+{
+public:
+        // Construct a wrapper around a default constructed object.
+        AIThreadSafeSimpleDC(void) { new (AIThreadSafeSimple<T>::ptr()) T; }
+
+protected:
+        // For use by AIThreadSafeSimpleDCRootPool
+        AIThreadSafeSimpleDC(AIAPRPool& parent) : AIThreadSafeSimple<T>(parent) { new (AIThreadSafeSimple<T>::ptr()) T; }
+};
+
+// Helper class for AIThreadSafeSimpleDCRootPool to assure initialization of
+// the root pool before constructing AIThreadSafeSimpleDC.
+class AIThreadSafeSimpleDCRootPool_pbase
+{
+protected:
+        AIAPRRootPool mRootPool;
+
+private:
+        template<typename T> friend class AIThreadSafeSimpleDCRootPool;
+        AIThreadSafeSimpleDCRootPool_pbase(void) { }
+};
+
+/**
+ * @brief A wrapper class for objects that need to be accessed by more than one thread.
+ *
+ * The same as AIThreadSafeSimpleDC except that this class creates its own AIAPRRootPool
+ * for the internally used mutexes and condition, instead of using the current threads
+ * root pool. The advantage of this is that it can be used for objects that need to
+ * be accessed from the destructors of global objects (after main). The disadvantage
+ * is that it's less efficient to use your own root pool, therefore it's use should be
+ * restricted to those cases where it is absolutely necessary.
+ */
+template<typename T>
+class AIThreadSafeSimpleDCRootPool : private AIThreadSafeSimpleDCRootPool_pbase, public AIThreadSafeSimpleDC<T>
+{
+public:
+        // Construct a wrapper around a default constructed object, using memory allocated
+        // from the operating system for the internal APR objects (mutexes and conditional),
+        // as opposed to allocated from the current threads root pool.
+        AIThreadSafeSimpleDCRootPool(void) :
+                AIThreadSafeSimpleDCRootPool_pbase(),
+                AIThreadSafeSimpleDC<T>(mRootPool) { }
+};
+
+/**
+ * @brief Write lock object and provide read/write access.
+ */
+template<typename T>
+struct AIAccess
+{
+        //! Construct a AIAccess from a non-constant AIThreadSafeSimple.
+        AIAccess(AIThreadSafeSimple<T>& wrapper) : mWrapper(wrapper) { this->mWrapper.mMutex.lock(); }
+
+        //! Access the underlaying object for (read and) write access.
+        T* operator->() const { return this->mWrapper.ptr(); }
+
+        //! Access the underlaying object for (read and) write access.
+        T& operator*() const { return *this->mWrapper.ptr(); }
+
+        ~AIAccess() { this->mWrapper.mMutex.unlock(); }
+
+protected:
+        AIThreadSafeSimple<T>& mWrapper;        //!< Reference to the object that we provide access to.
+
+private:
+        // Disallow copy constructing directly.
+        AIAccess(AIAccess const&);
+};
+
+#endif
diff --git a/linden/indra/llcommon/llthread.h b/linden/indra/llcommon/llthread.h
index 8708929..1e982cc 100644
--- a/linden/indra/llcommon/llthread.h
+++ b/linden/indra/llcommon/llthread.h
@@ -235,6 +235,101 @@ private:
         LLMutexBase* mMutex;
 };
 
+class AIRWLock
+{
+public:
+        AIRWLock(AIAPRPool& parent = LLThread::tldata().mRootPool) :
+                mWriterWaitingMutex(parent), mNoHoldersCondition(parent), mHoldersCount(0), mWriterIsWaiting(false) { }
+
+private:
+        LLMutex mWriterWaitingMutex;            //!< This mutex is locked while some writer is waiting for access.
+        LLCondition mNoHoldersCondition;        //!< Access control for mHoldersCount. Condition true when there are no more holders.
+        int mHoldersCount;                                      //!< Number of readers or -1 if a writer locked this object.
+        // This is volatile because we read it outside the critical area of mWriterWaitingMutex, at [1].
+        // That means that other threads can change it while we are already in the (inlined) function rdlock.
+        // Without volatile, the following assembly would fail:
+        // register x = mWriterIsWaiting;
+        // /* some thread changes mWriterIsWaiting */
+        // if (x ...
+        // However, because the function is fuzzy to begin with (we don't mind that this race
+        // condition exists) it would work fine without volatile. So, basically it's just here
+        // out of principle ;). -- Aleric
+    bool volatile mWriterIsWaiting;             //!< True when there is a writer waiting for write access.
+
+public:
+        void rdlock(bool high_priority = false)
+        {
+                // Give a writer a higher priority (kinda fuzzy).
+                if (mWriterIsWaiting && !high_priority) // [1] If there is a writer interested,
+                {
+                        mWriterWaitingMutex.lock();                     // [2] then give it precedence and wait here.
+                        // If we get here then the writer got it's access; mHoldersCount == -1.
+                        mWriterWaitingMutex.unlock();
+                }
+                mNoHoldersCondition.lock();                             // [3] Get exclusive access to mHoldersCount.
+                while (mHoldersCount == -1)                             // [4]
+                {
+                        mNoHoldersCondition.wait();                     // [5] Wait till mHoldersCount is (or just was) 0.
+                }
+                ++mHoldersCount;                                                // One more reader.
+                mNoHoldersCondition.unlock();                   // Release lock on mHoldersCount.
+        }
+        void rdunlock(void)
+        {
+                mNoHoldersCondition.lock();                             // Get exclusive access to mHoldersCount.
+                if (--mHoldersCount == 0)                               // Was this the last reader?
+                {
+                        mNoHoldersCondition.signal();           // Tell waiting threads, see [5], [6] and [7].
+                }
+                mNoHoldersCondition.unlock();                   // Release lock on mHoldersCount.
+        }
+        void wrlock(void)
+        {
+                mWriterWaitingMutex.lock();                             // Block new readers, see [2],
+                mWriterIsWaiting = true;                                // from this moment on, see [1].
+                mNoHoldersCondition.lock();                             // Get exclusive access to mHoldersCount.
+                while (mHoldersCount != 0)                              // Other readers or writers have this lock?
+                {
+                        mNoHoldersCondition.wait();                     // [6] Wait till mHoldersCount is (or just was) 0.
+                }
+                mWriterIsWaiting = false;                               // Stop checking the lock for new readers, see [1].
+                mWriterWaitingMutex.unlock();                   // Release blocked readers, they will still hang at [3].
+                mHoldersCount = -1;                                             // We are a writer now (will cause a hang at [5], see [4]).
+                mNoHoldersCondition.unlock();                   // Release lock on mHolders (readers go from [3] to [5]).
+        }
+        void wrunlock(void)
+        {
+                mNoHoldersCondition.lock();                             // Get exclusive access to mHoldersCount.
+                mHoldersCount = 0;                                              // We have no writer anymore.
+                mNoHoldersCondition.signal();                   // Tell waiting threads, see [5], [6] and [7].
+                mNoHoldersCondition.unlock();                   // Release lock on mHoldersCount.
+        }
+        void rd2wrlock(void)
+        {
+                mNoHoldersCondition.lock();                             // Get exclusive access to mHoldersCount. Blocks new readers at [3].
+                if (--mHoldersCount > 0)                                // Any other reads left?
+                {
+                        mWriterWaitingMutex.lock();                     // Block new readers, see [2],
+                        mWriterIsWaiting = true;                        // from this moment on, see [1].
+                        while (mHoldersCount != 0)                      // Other readers (still) have this lock?
+                        {
+                                mNoHoldersCondition.wait();             // [7] Wait till mHoldersCount is (or just was) 0.
+                        }
+                        mWriterIsWaiting = false;                       // Stop checking the lock for new readers, see [1].
+                        mWriterWaitingMutex.unlock();           // Release blocked readers, they will still hang at [3].
+                }
+                mHoldersCount = -1;                                             // We are a writer now (will cause a hang at [5], see [4]).
+                mNoHoldersCondition.unlock();                   // Release lock on mHolders (readers go from [3] to [5]).
+        }
+        void wr2rdlock(void)
+        {
+                mNoHoldersCondition.lock();                             // Get exclusive access to mHoldersCount.
+                mHoldersCount = 1;                                              // Turn writer into a reader.
+                mNoHoldersCondition.signal();                   // Tell waiting readers, see [5].
+                mNoHoldersCondition.unlock();                   // Release lock on mHoldersCount.
+        }
+};
+
 //============================================================================
 
 void LLThread::lockData()
@@ -247,7 +342,6 @@ void LLThread::unlockData()
         mRunCondition->unlock();
 }
 
-
 //============================================================================
 
 // see llmemory.h for LLPointer<> definition