BVhHdZddlmZddlmZddlmZdZGddeZ y) zMemory leak detection utility.)_PythonMemoryChecker)trace) tf_inspectctjD]I}|d}|jds |djdjj }|dz|zcSy#YQxYw)z>If available, return the current test name. Otherwise, `None`.testrself.N)rstack startswithf_locals __class____name__)r function_name class_names Z/home/dcms/DCMS/lib/python3.12/site-packages/tensorflow/python/framework/memory_checker.py_get_test_name_best_effortrsu! e!HM' 1X&&v.88AA C-//     s -A""A&ceZdZdZej dZej dZdZej dZ ej dZ ej d dZ y) MemoryCheckeraMemory leak detection class. This is a utility class to detect Python and C++ memory leaks. It's intended for both testing and debugging. Basic usage: >>> # MemoryChecker() context manager tracks memory status inside its scope. >>> with MemoryChecker() as memory_checker: >>> tensors = [] >>> for _ in range(10): >>> # Simulating `tf.constant(1)` object leak every iteration. >>> tensors.append(tf.constant(1)) >>> >>> # Take a memory snapshot for later analysis. >>> memory_checker.record_snapshot() >>> >>> # `report()` generates a html graph file showing allocations over >>> # snapshots per every stack trace. >>> memory_checker.report() >>> >>> # This assertion will detect `tf.constant(1)` object leak. >>> memory_checker.assert_no_leak_if_all_possibly_except_one() `record_snapshot()` must be called once every iteration at the same location. This is because the detection algorithm relies on the assumption that if there is a leak, it's happening similarly on every snapshot. c$t|_|SN)r_python_memory_checkerr s r __enter__zMemoryChecker.__enter__As"6"8D Kcyr)r exc_type exc_value tracebacks r__exit__zMemoryChecker.__exit__Fsrc8|jjy)aTake a memory snapshot for later analysis. `record_snapshot()` must be called once every iteration at the same location. This is because the detection algorithm relies on the assumption that if there is a leak, it's happening similarly on every snapshot. The recommended number of `record_snapshot()` call depends on the testing code complexity and the allcoation pattern. N)rrecord_snapshotrs rr#zMemoryChecker.record_snapshotLs //1rc8|jjy)aGenerates a html graph file showing allocations over snapshots. It create a temporary directory and put all the output files there. If this is running under Google internal testing infra, it will use the directory provided the infra instead. N)rreportrs rr%zMemoryChecker.reportXs &&(rc8|jjy)zRaises an exception if a leak is detected. This algorithm classifies a series of allocations as a leak if it's the same type(Python) or it happens at the same stack trace(C++) at every snapshot, but possibly except one snapshot. N)r)assert_no_leak_if_all_possibly_except_oners rr'z7MemoryChecker.assert_no_leak_if_all_possibly_except_onebs IIKrNc<|jj|y)aNRaises an exception if there are new Python objects created. It computes the number of new Python objects per type using the first and the last snapshots. Args: threshold: A dictionary of [Type name string], [count] pair. It won't raise an exception if the new Python objects are under this threshold. ) thresholdN)rassert_no_new_objects)r r)s rassert_no_new_python_objectsz*MemoryChecker.assert_no_new_python_objectsms 55 5Jrr) r __module__ __qualname____doc__r trace_wrapperrr!r#r%r'r+rrrrr%s6    2 )) LL  K KrrN) r.1tensorflow.python.framework.python_memory_checkerrtensorflow.python.profilerrtensorflow.python.utilrrobjectrrrrr4s(%R,- SKFSKr