dict: Rework reference counting

Use C11 atomics if available. Otherwise, fall back to pthread mutex
inlined in xmlDict struct or Win32 atomics.

This should fix lock contention reported in #970.

include/private/dict.h

@@ -50,11 +50,6 @@ typedef struct {
     const xmlChar *name;
 } xmlHashedString;
 
-XML_HIDDEN void
-xmlInitDictInternal(void);
-XML_HIDDEN void
-xmlCleanupDictInternal(void);
-
 XML_HIDDEN unsigned
 xmlDictComputeHash(const xmlDict *dict, const xmlChar *string);
 XML_HIDDEN unsigned

include/private/threads.h

@@ -48,6 +48,141 @@ struct _xmlRMutex {
 #endif
 };
 
+#if defined(LIBXML_THREAD_ENABLED) && \
+    __STDC_VERSION__ >= 201112L && !defined(__STDC_NO_ATOMICS__)
+
+/** Atomic reference count */
+typedef _Atomic size_t xmlRefCount;
+
+/**
+ * Initialize refcount.
+ *
+ * @param r  refcount
+ */
+static XML_INLINE void
+xmlRefCountInit(xmlRefCount *r) {
+    *r = 1;
+}
+
+/**
+ * Increase refcount.
+ *
+ * @param r  refcount
+ */
+static XML_INLINE void
+xmlRefCountInc(xmlRefCount *r) {
+    *r += 1;
+}
+
+/**
+ * Decrease refcount.
+ *
+ * @param r  refcount
+ * @returns 0 if refcount reached zero, 1 otherwise
+ */
+static XML_INLINE int
+xmlRefCountDec(xmlRefCount *r) {
+    return --*r > 0;
+}
+
+#elif defined(HAVE_POSIX_THREADS)
+
+typedef struct {
+    pthread_mutex_t mutex;
+    size_t count;
+} xmlRefCount;
+
+static XML_INLINE void
+xmlRefCountInit(xmlRefCount *r) {
+    pthread_mutex_init(&r->mutex, NULL);
+    r->count = 1;
+}
+
+static XML_INLINE void
+xmlRefCountInc(xmlRefCount *r) {
+    pthread_mutex_lock(&r->mutex);
+    r->count += 1;
+    pthread_mutex_unlock(&r->mutex);
+}
+
+static XML_INLINE int
+xmlRefCountDec(xmlRefCount *r) {
+    size_t val;
+
+    pthread_mutex_lock(&r->mutex);
+    val = --r->count;
+    pthread_mutex_unlock(&r->mutex);
+
+    if (val > 0)
+        return 1;
+
+    pthread_mutex_destroy(&r->mutex);
+    return 0;
+}
+
+#elif defined(HAVE_WIN32_THREADS)
+
+#ifdef _WIN64
+
+typedef __int64 xmlRefCount;
+
+static XML_INLINE void
+xmlRefCountInit(xmlRefCount *r) {
+    *r = 1;
+}
+
+static XML_INLINE void
+xmlRefCountInc(xmlRefCount *r) {
+    InterlockedIncrement64(r);
+}
+
+static XML_INLINE int
+xmlRefCountDec(xmlRefCount *r) {
+    return InterlockedDecrement64(r) > 0;
+}
+
+#else /* 32-bit */
+
+typedef long xmlRefCount;
+
+static XML_INLINE void
+xmlRefCountInit(xmlRefCount *r) {
+    *r = 1;
+}
+
+static XML_INLINE void
+xmlRefCountInc(xmlRefCount *r) {
+    InterlockedIncrement(r);
+}
+
+static XML_INLINE int
+xmlRefCountDec(xmlRefCount *r) {
+    return InterlockedDecrement(r) > 0;
+}
+
+#endif
+
+#else /* no threads */
+
+typedef size_t xmlRefCount;
+
+static XML_INLINE void
+xmlRefCountInit(xmlRefCount *r) {
+    *r = 1;
+}
+
+static XML_INLINE void
+xmlRefCountInc(xmlRefCount *r) {
+    *r += 1;
+}
+
+static XML_INLINE int
+xmlRefCountDec(xmlRefCount *r) {
+    return --*r > 0;
+}
+
+#endif
+
 XML_HIDDEN void
 xmlInitMutex(xmlMutex *mutex);
 XML_HIDDEN void