Prepare for RefCounting Anonymous Variables.

The documented case is no longer relevant. After cl/401590080, tf.function captured anonymous variables hold strong reference to the variables via the handle tensors.

Shall be submitted together with cl/401590080.

PiperOrigin-RevId: 406882396

Author: tensorflower-gardener

site/en/guide/function.ipynb

@@ -1167,43 +1167,6 @@
         "good_consume_next(iterator)"
       ]
     },
-    {
-      "cell_type": "markdown",
-      "metadata": {
-        "id": "FHQ0UeU-vWo8"
-      },
-      "source": [
-        "### Deleting tf.Variables between `Function` calls\n",
-        "\n",
-        "Another error you may encounter is a garbage-collected variable. `ConcreteFunction`s only retain [WeakRefs](https://docs.python.org/3/library/weakref.html) to the variables they close over, so you must retain a reference to any variables."
-      ]
-    },
-    {
-      "cell_type": "code",
-      "execution_count": null,
-      "metadata": {
-        "id": "uMiRPfETjpt-"
-      },
-      "outputs": [],
-      "source": [
-        "external_var = tf.Variable(3)\n",
-        "@tf.function\n",
-        "def f(x):\n",
-        "  return x * external_var\n",
-        "\n",
-        "traced_f = f.get_concrete_function(4)\n",
-        "print(\"Calling concrete function...\")\n",
-        "print(traced_f(4))\n",
-        "\n",
-        "# The original variable object gets garbage collected, since there are no more\n",
-        "# references to it.\n",
-        "external_var = tf.Variable(4)\n",
-        "print()\n",
-        "print(\"Calling concrete function after garbage collecting its closed Variable...\")\n",
-        "with assert_raises(tf.errors.FailedPreconditionError):\n",
-        "  traced_f(4)"
-      ]
-    },
     {
       "cell_type": "markdown",
       "metadata": {
@@ -1379,10 +1342,39 @@
     {
       "cell_type": "markdown",
       "metadata": {
-        "id": "Tu0SnPwaL7pI"
+        "id": "ZoPg5w1Pjqnb"
       },
       "source": [
-        "You can close over outer names, as long as you don't update their values.\n"
+        "Another way to update a global value, is to make it a `tf.Variable` and use the `Variable.assign` method instead.\n"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {
+        "id": "oeJMdXd3M0cc"
+      },
+      "outputs": [],
+      "source": [
+        "@tf.function\n",
+        "def variable_add():\n",
+        "  return 1 + foo\n",
+        "\n",
+        "foo = tf.Variable(1)\n",
+        "print(\"Variable:\", variable_add())\n"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {
+        "id": "L3q7sUJWZOSd"
+      },
+      "outputs": [],
+      "source": [
+        "print(\"Updating the value of `foo` to 100!\")\n",
+        "foo.assign(100)\n",
+        "print(\"Variable:\", variable_add())"
       ]
     },
     {