From: magnus.blaudd
Date: October 17 2012 1:27pm
Subject: bzr push into mysql-5.1-telco-7.1 branch (magnus.blaudd:4628 to 4629)
List-Archive: http://lists.mysql.com/commits/145071
Message-Id: <20121017132730.31943.49170.4629@wholphin>
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit

 4629 magnus.blaudd@stripped	2012-10-17 [merge]
      Merge 7.0 -> 7.1

    modified:
      storage/ndb/src/common/portlib/NdbThread.c
 4628 magnus.blaudd@stripped	2012-10-17 [merge]
      Merge

    modified:
      mysql-test/lib/My/Exec.pm
=== modified file 'storage/ndb/src/common/portlib/NdbThread.c'
--- a/storage/ndb/src/common/portlib/NdbThread.c	2011-11-10 15:06:03 +0000
+++ b/storage/ndb/src/common/portlib/NdbThread.c	2012-10-17 13:18:33 +0000
@@ -61,6 +61,25 @@ struct NdbThread
 { 
   volatile int inited;
   pthread_t thread;
+#ifdef _WIN32
+  /*
+    Problem in mysys on certain MySQL versions where the thread id is
+    used as the thread identifier. Since the thread id may be reused
+    by another thread(when the current thread exits) one may end up
+    waiting for the wrong thread.
+
+    Workaround by using a HANDLE which is opened by thread itself in
+    'ndb_thread_wrapper' and subsequently used to wait for the
+    thread in 'NdbThread_WaitFor'.
+
+    NOTE: Windows implementation of 'pthread_join' and support for non
+    detached threads as implemented in MySQL Cluster 7.0 and 7.1 is not
+    affected, only MySQL Cluster based on 5.5+ affected(slightly
+    different implementation). The current workaround in NdbThread is
+    generic and works regardless of pthread_join implementation in mysys.
+  */
+  HANDLE thread_handle;
+#endif
 #if defined HAVE_SOLARIS_AFFINITY
   id_t tid;
 #elif defined HAVE_LINUX_SCHEDULING
@@ -149,6 +168,15 @@ ndb_thread_wrapper(void* _ss){
       struct NdbThread * ss = (struct NdbThread *)_ss;
       settid(ss);
 
+#ifdef _WIN32
+      /*
+        Create the thread handle, ignore failure since it's unlikely
+        to fail and the functions using this handle checks for NULL.
+      */
+      ss->thread_handle =
+        OpenThread(SYNCHRONIZE, FALSE, GetCurrentThreadId());
+#endif
+
 #ifdef NDB_MUTEX_DEADLOCK_DETECTOR
       ndb_mutex_thread_init(&ss->m_mutex_thr_state);
 #endif
@@ -313,6 +341,11 @@ void NdbThread_Destroy(struct NdbThread*
   DBUG_ENTER("NdbThread_Destroy");
   if (*p_thread != NULL){
     DBUG_PRINT("enter",("*p_thread: 0x%lx", (long) *p_thread));
+#ifdef _WIN32
+    HANDLE thread_handle = (*p_thread)->thread_handle;
+    if (thread_handle)
+      CloseHandle(thread_handle);
+#endif
     free(* p_thread); 
     * p_thread = 0;
   }
@@ -322,17 +355,35 @@ void NdbThread_Destroy(struct NdbThread*
 
 int NdbThread_WaitFor(struct NdbThread* p_wait_thread, void** status)
 {
-  int result;
-
   if (p_wait_thread == NULL)
     return 0;
 
   if (p_wait_thread->thread == 0)
     return 0;
 
-  result = pthread_join(p_wait_thread->thread, status);
-  
-  return result;
+#ifdef _WIN32
+  {
+    DWORD ret;
+    HANDLE thread_handle = p_wait_thread->thread_handle;
+
+    if (thread_handle == NULL)
+    {
+      return -1;
+    }
+
+    ret = WaitForSingleObject(thread_handle, INFINITE);
+    if (ret != WAIT_OBJECT_0)
+    {
+      return -1;
+    }
+
+    /* Don't fill in "status", never used anyway */
+
+    return 0;
+  }
+#else
+  return pthread_join(p_wait_thread->thread, status);
+#endif
 }
 
 

No bundle (reason: useless for push emails).