Function dbm_length from Modules/_gdbmmodule.c seems to be leaking memory. A simple reproducer: import dbm d = dbm.open('spam', 'c') d['x'] = '1' print(len(d)) The interesting part of valgrind output with --leak-check=full: ==3312== 1 bytes in 1 blocks are definitely lost in loss record 1 of 3,622 ==3312== at 0x4A06409: malloc (in /usr/lib64/valgrind/vgpreload_memcheck-amd64-linux.so) ==3312== by 0x3A31A0349D: get_next_key (gdbmseq.c:72) ==3312== by 0x3A31A03595: gdbm_nextkey (gdbmseq.c:126) ==3312== by 0xDD74192: dbm_length (_gdbmmodule.c:104) ==3312== by 0x3A2473F50B: builtin_len (bltinmodule.c:1293) ==3312== by 0x3A247510D1: PyEval_EvalFrameEx (ceval.c:4080) ==3312== by 0x3A24753758: PyEval_EvalCodeEx (ceval.c:3462) ==3312== by 0x3A2475390A: PyEval_EvalCode (ceval.c:791) ==3312== by 0x3A24773893: run_mod (pythonrun.c:1989) ==3312== by 0x3A24775EE7: PyRun_FileExFlags (pythonrun.c:1945) ==3312== by 0x3A24777000: PyRun_SimpleFileExFlags (pythonrun.c:1455) ==3312== by 0x3A2478F72B: Py_Main (main.c:306) It seems that the problem is in the loop here [1] - specifically, the last okey.dptr seems not to be freed (the loop is terminated when key.dptr is NULL, but okey.dptr is still pointing to something). I'm attaching a supersimple patch that should fix this (unless I'm totally wrong about everything...) [1] http://hg.python.org/cpython/file/65f2c92ed079/Modules/_gdbmmodule.c#l102
I just tried rebuilding with my patch and running test suite + I did some manual testing and everything seems to work fine - and the memory leak is not there.
