This document describes some caveats about the use of Valgrind with Python. Valgrind is used periodically by Python developers to try to ensure there are no memory leaks or invalid memory reads/writes. If you don't want to read about the details of using Valgrind, there are still two things you must do to suppress the warnings. First, you must use a suppressions file. One is supplied in Misc/valgrind-python.supp. Second, you must do one of the following: * Uncomment Py_USING_MEMORY_DEBUGGER in Objects/obmalloc.c, then rebuild Python * Uncomment the lines in Misc/valgrind-python.supp that suppress the warnings for PyObject_Free and PyObject_Realloc If you want to use Valgrind more effectively and catch even more memory leaks, you will need to configure python --without-pymalloc. PyMalloc allocates a few blocks in big chunks and most object allocations don't call malloc, they use chunks doled about by PyMalloc from the big blocks. This means Valgrind can't detect many allocations (and frees), except for those that are forwarded to the system malloc. Note: configuring python --without-pymalloc makes Python run much slower, especially when running under Valgrind. You may need to run the tests in batches under Valgrind to keep the memory usage down to allow the tests to complete. It seems to take about 5 times longer to run --without-pymalloc. Apr 15, 2006: test_ctypes causes Valgrind 3.1.1 to fail (crash). test_socket_ssl should be skipped when running valgrind. The reason is that it purposely uses uninitialized memory. This causes many spurious warnings, so it's easier to just skip it. Details: -------- Python uses its own small-object allocation scheme on top of malloc, called PyMalloc. Valgrind may show some unexpected results when PyMalloc is used. Starting with Python 2.3, PyMalloc is used by default. You can disable PyMalloc when configuring python by adding the --without-pymalloc option. If you disable PyMalloc, most of the information in this document and the supplied suppressions file will not be useful. As discussed above, disabling PyMalloc can catch more problems. If you use valgrind on a default build of Python, you will see many errors like: ==6399== Use of uninitialised value of size 4 ==6399== at 0x4A9BDE7E: PyObject_Free (obmalloc.c:711) ==6399== by 0x4A9B8198: dictresize (dictobject.c:477) These are expected and not a problem. Tim Peters explains the situation: PyMalloc needs to know whether an arbitrary address is one that's managed by it, or is managed by the system malloc. The current scheme allows this to be determined in constant time, regardless of how many memory areas are under pymalloc's control. The memory pymalloc manages itself is in one or more "arenas", each a large contiguous memory area obtained from malloc. The base address of each arena is saved by pymalloc in a vector. Each arena is carved into "pools", and a field at the start of each pool contains the index of that pool's arena's base address in that vector. Given an arbitrary address, pymalloc computes the pool base address corresponding to it, then looks at "the index" stored near there. If the index read up is out of bounds for the vector of arena base addresses pymalloc maintains, then pymalloc knows for certain that this address is not under pymalloc's control. Otherwise the index is in bounds, and pymalloc compares the arena base address stored at that index in the vector to the arbitrary address pymalloc is investigating pymalloc controls this arbitrary address if and only if it lies in the arena the address's pool's index claims it lies in. It doesn't matter whether the memory pymalloc reads up ("the index") is initialized. If it's not initialized, then whatever trash gets read up will lead pymalloc to conclude (correctly) that the address isn't controlled by it, either because the index is out of bounds, or the index is in bounds but the arena it represents doesn't contain the address. This determination has to be made on every call to one of pymalloc's free/realloc entry points, so its speed is critical (Python allocates and frees dynamic memory at a ferocious rate -- everything in Python, from integers to "stack frames", lives in the heap).