examine and deliberate security concerns

[bertsky]

trafficstars

As long as we only have command-line interfaces, I suppose security is a matter of system administration (user priviledges for FS access and ulimit restriction of runtime resources). It's difficult to define and track what is a fair resource consumption. But as soon as we integrate them into network services (as scripts or by directly adding network capabilities in the form of low-level workflow/processing servers or a Web API implementation), we will also need to protect users from each other and against malign input.

For example, most of our XML processing is done via lxml (based on libxml2), which comes off rather bad regarding resource overconsumption: https://pypi.org/project/defusedxml. (The input METS or PAGE could be compromised in order to DOS-attack the machine or leak sensitive information via XInclude directives.)

To my knowledge, no analysis of such potential problems has been undertaken for OCR-D yet.