MTN deterministic co_filename for dynamic code pickling

cloudpickle/cloudpickle.py

@@ -829,6 +829,11 @@ def _code_reduce(obj):
     # See the inline comment in _class_setstate for details.
     co_name = "".join(obj.co_name)
 
+    # co_filename is not used in the constructor of code objects, so we can
+    # safely set it to indicate that this is dynamic code. This also makes
+    # the payload deterministic, independent of where the function is defined.
+    co_filename = "".join("<dynamic-code>")
+
     # Create shallow copies of these tuple to make cloudpickle payload deterministic.
     # When creating a code object during load, copies of these four tuples are
     # created, while in the main process, these tuples can be shared.
@@ -851,7 +856,7 @@ def _code_reduce(obj):
             obj.co_consts,
             co_names,
             co_varnames,
-            obj.co_filename,
+            co_filename,
             co_name,
             obj.co_qualname,
             obj.co_firstlineno,
@@ -874,7 +879,7 @@ def _code_reduce(obj):
             obj.co_consts,
             co_names,
             co_varnames,
-            obj.co_filename,
+            co_filename,
             co_name,
             obj.co_firstlineno,
             obj.co_linetable,
@@ -895,7 +900,7 @@ def _code_reduce(obj):
             obj.co_code,
             obj.co_consts,
             co_varnames,
-            obj.co_filename,
+            co_filename,
             co_name,
             obj.co_firstlineno,
             obj.co_lnotab,
@@ -919,7 +924,7 @@ def _code_reduce(obj):
             obj.co_consts,
             co_names,
             co_varnames,
-            obj.co_filename,
+            co_filename,
             co_name,
             obj.co_firstlineno,
             obj.co_lnotab,

dev-requirements.txt

@@ -7,6 +7,8 @@ pytest-cov
 psutil
 # To be able to test tornado coroutines
 tornado
+# To be able to test behavior in jupyter-notbooks
+ipykernel
 # To be able to test numpy specific things
 # but do not build numpy from source on Python nightly
 numpy >=1.18.5; python_version <= '3.12'

tests/cloudpickle_ipykernel_test.py

@@ -0,0 +1,64 @@
+import pytest
+import textwrap
+
+from .testutils import check_deterministic_pickle
+
+try:
+    import ipykernel
+except ImportError:
+    pytest.skip("ipykernel is not installed", allow_module_level=True)
+
+
+def run_in_notebook(code):
+    km = ipykernel.connect.jupyter_client.KernelManager()
+    km.start_kernel()
+    kc = km.client()
+    kc.start_channels()
+    status, output, err = "run", None, None
+    try:
+        assert km.is_alive() and kc.is_alive()
+        kc.wait_for_ready()
+        idx = kc.execute(code)
+        running = True
+        while running:
+            res = kc.iopub_channel.get_msg(timeout=None)
+            if res['parent_header'].get('msg_id') != idx:
+                continue
+            content = res['content']
+            if content.get("name", "state") == "stdout":
+                output = content['text']
+            if "traceback" in content:
+                err = "\n".join(content['traceback'])
+                status = "error"
+            running = res['content'].get('execution_state', None) != "idle"
+    finally:
+        kc.shutdown()
+        kc.stop_channels()
+        km.shutdown_kernel(now=True, restart=False)
+        assert not km.is_alive()
+    if status != "error":
+        status = "ok" if not running else "exec_error"
+    return status, output, err
+
+
+def test_deterministic_payload_for_dynamic_func_in_notebook():
+    code = textwrap.dedent("""
+        import cloudpickle
+
+        MY_PI = 3.1415
+
+        def get_pi():
+            return MY_PI
+
+        print(cloudpickle.dumps(get_pi))
+    """)
+
+    status, output, err = run_in_notebook(code)
+    assert status == "ok"
+    payload = eval(output.strip(), {})
+
+    status, output, err = run_in_notebook(code)
+    assert status == "ok"
+    payload2 = eval(output.strip(), {})
+
+    check_deterministic_pickle(payload, payload2)