It looks like that's the way to do it : http://modules.apache.org/doc/API.html#per-dir http://www.apachetutor.org/dev/config
But the memory management seems very mysterious to me here... I can't find anything that looks like a destructor for the py_config structure ! Regards, Nicolas 2005/6/24, Jim Gallacher <[EMAIL PROTECTED]>: > Anybody had a chance to look at this yet? > > I'm kind of suspicous of python_merge_config in mod_python.c. It gets > called once for every request, and calls apr_pcalloc which I assume > allocates some memory from the pool. > > py_config *merged_conf = > (py_config *) apr_pcalloc(p, sizeof(py_config)); > > I don't see where or how this memory gets freed. I've taken a look at > the apache src code but get lost pretty quick. > > Regards, > Jim > > > Jim Gallacher (JIRA) wrote: > > PythonOption directive causes memory leak > > ----------------------------------------- > > > > Key: MODPYTHON-60 > > URL: http://issues.apache.org/jira/browse/MODPYTHON-60 > > Project: mod_python > > Type: Bug > > Components: core > > Versions: 3.1.4, 3.1.3, 3.2.0 > > Environment: Linux > > Reporter: Jim Gallacher > > Priority: Critical > > > > > > This was previously reported on the mod_python mailing list. See > > http://www.modpython.org/pipermail/mod_python/2004-April/015395.html > > > > A memory leak results when there is a PythonOption directive in the apache > > config file. Leak occurs when PythonOption is in either <VirtualHost> or > > <Directory> section. > > > > For each request, approx 25 bytes of memory is leaked per PythonOption > > directive. > > > > Methodolgy (using top to gauge memory usage, 100,000 requests per test > > case): > > > > def handler(req): > > req.content_type = 'text/plain' > > req.write('PythonOption test\n') > > return apache.OK > > > > 1. No PythonOption directives: > > 1.4 % MEM > > > > 2. 50 PythonOption directives: > > 11.3% MEM > > > > 3. 100 PythonOption directives: > > 25.4 % MEM > > > > > > I know 50 or 100 PythonOptions is not likely in a production system, but it > > clearly demonstrate the leak. > > > > > > > >
