diff --git a/compiler/lowerings.nim b/compiler/lowerings.nim
index df2816a0e..2fff4a6f1 100644
--- a/compiler/lowerings.nim
+++ b/compiler/lowerings.nim
@@ -205,7 +205,7 @@ proc createNimCreatePromiseCall(prom, threadParam: PNode): PNode =
   let castExpr = newNodeIT(nkCast, prom.info, prom.typ)
   castExpr.add emptyNode
   castExpr.add callCodeGenProc("nimCreatePromise", threadParam, size)
-  result = newFastAsgnStmt(prom, castExpr)
+  result = castExpr
 
 proc createWrapperProc(f: PNode; threadParam, argsParam: PSym;
                        varSection, call, barrier, prom: PNode;
@@ -216,9 +216,14 @@ proc createWrapperProc(f: PNode; threadParam, argsParam: PSym;
   var threadLocalProm: PSym
   if spawnKind == srByVar:
     threadLocalProm = addLocalVar(varSection, argsParam.owner, prom.typ, prom)
+  elif prom != nil:
+    internalAssert prom.typ.kind == tyGenericInst
+    threadLocalProm = addLocalVar(varSection, argsParam.owner, prom.typ, 
+      createNimCreatePromiseCall(prom, threadParam.newSymNode))
+    
   body.add varSection
   if prom != nil and spawnKind != srByVar:
-    body.add createNimCreatePromiseCall(prom, threadParam.newSymNode)
+    body.add newFastAsgnStmt(prom, threadLocalProm.newSymNode)
     if barrier == nil:
       body.add callCodeGenProc("nimPromiseCreateCondVar", prom)
 
@@ -230,10 +235,12 @@ proc createWrapperProc(f: PNode; threadParam, argsParam: PSym;
     if fk == promInvalid:
       localError(f.info, "cannot create a promise of type: " & 
         typeToString(prom.typ.sons[1]))
-    body.add newAsgnStmt(indirectAccess(prom,
+    body.add newAsgnStmt(indirectAccess(threadLocalProm.newSymNode,
       if fk == promGC: "data" else: "blob", prom.info), call)
     if barrier == nil:
-      body.add callCodeGenProc("nimPromiseSignal", prom)
+      # by now 'prom' is shared and thus might have beeen overwritten! we need
+      # to use the thread-local view instead:
+      body.add callCodeGenProc("nimPromiseSignal", threadLocalProm.newSymNode)
   else:
     body.add call
   if barrier != nil:
diff --git a/lib/pure/concurrency/threadpool.nim b/lib/pure/concurrency/threadpool.nim
index 24cb9ccdd..3a7693791 100644
--- a/lib/pure/concurrency/threadpool.nim
+++ b/lib/pure/concurrency/threadpool.nim
@@ -151,7 +151,7 @@ proc await*[T](prom: Promise[T]) =
   if prom.usesCondVar: await(prom.cv)
 
 proc awaitAndThen*[T](prom: Promise[T]; action: proc (x: T) {.closure.}) =
-  ## blocks until the value is available and then passes this value
+  ## blocks until the ``prom`` is available and then passes its value
   ## to ``action``. Note that due to Nimrod's parameter passing semantics this
   ## means that ``T`` doesn't need to be copied and so ``awaitAndThen`` can
   ## sometimes be more efficient than ``^``.
diff --git a/tests/parallel/nimrod.cfg b/tests/parallel/nimrod.cfg
new file mode 100644
index 000000000..b81c89721
--- /dev/null
+++ b/tests/parallel/nimrod.cfg
@@ -0,0 +1 @@
+threads:on