when use xml.dom.pulldom module to parse a large xml file, if all the
information is saved in one xml file, the module can handle it in the
following way without construction the whole DOM:
events = xml.dom.pulldom.parse('file.xml')
for (event, node) in events:
process(event, node)
But if 'file.xml' contains some large external entities, for example:
<!ENTITY file_external SYSTEM "others.xml">
<body>&file_external;</body>
Then using the same python snippet above leads to enormous memory
usage. I did not perform a concrete benchmark, in one case a 3M
external xml file drained about 1 GB memory. I think in this case it
might be the whole DOM structure is constructed.