I'm attaching a patch which generalizes the at-exit PYTHONMALLOCSTATS memory usage report, so that it's available in a regular build and can be triggered from Python, by calling:
   sys._debugmallocstats()

This can be useful when debugging memory usage issues: a script can call the debug hook before and after certain activities, and the before/after amounts can be compared.

The patch moves this arena-accouting code when a new arena is allocated:
     ++ntimes_arena_allocated;
     if (narenas_currently_allocated > narenas_highwater)
         narenas_highwater = narenas_currently_allocated;
from out of the #ifdef PYMALLOC_DEBUG guard and into all PYMALLOC configurations, so that this data is available within the dump.  Given how much activity happens when a new arena is allocated, I believe this doesn't impact performance.

It changes _PyObject_DebugMallocStats() to take an arbitrary FILE*, rather than assuming stderr (which was handy for my original use-case of debugging a web server).  This function is already marked with PyAPI_FUNC() but not documented (albeit only present in a debug build).

Tested with --with-pymalloc, --without-pymalloc, and --with-pydebug

FWIW, Red Hat has been using a version of this patch in RHEL 5 as of RHEL 5.6 (http://rhn.redhat.com/errata/RHSA-2011-0027.html), and also in Fedora since September 2011 with python-2.7.2-15 and python3-3.2.2-6 (for the forthcoming Fedora 17).

Nice idea. I don't see any obvious problem with the patch, except that the test should reuse test.script_helper.assert_python_ok().

Updated version of the patch, using test.script_helper.assert_python_ok() and adding a NEWS entry

One other nit: the C API functions shouldn't be included in the limited API.

Thanks.  I'm attaching an updated version of the patch, wrapping all new C entrypoints within a #ifndef Py_LIMITED_API

I also moved the existing _PyObject_DebugMallocStats() entrypoint to within a #ifndef Py_LIMITED_API.  As noted above, it is not documented (and these patches change it from accepting void to a FILE*).

New changeset d63a80abfbec by David Malcolm in branch 'default':
Issue #14785: Add sys._debugmallocstats() to help debug low-level memory allocation issues
http://hg.python.org/cpython/rev/d63a80abfbec