XML must die
If I have been quiet over the last few days, that's mostly because I have been busy designing a concept for the Personal UPload Server (PUPS). It is a system that runs on an HTTP server and allows an app to get credentials for uploading files to that HTTP server at predefined locations. This will give apps the capability to create and post to feeds, and to update SocialSpec profiles.
I was critical of using XML-based formats like RSS for a while, because they are difficult to use correctly. First, you need an XML library, which by itself is not that easy today on less popular platforms. Then you need to use it correctly - not all libraries make it easy to handle all the less-obvious features like CDATA, entities, comments, and namespaces. And you must resist the urge to write a couple of regexes to avoid all that, because they will always fail in XML edge cases.
XML is getting worse when you need to edit XML files, especially those that are also edited by humans. It's almost impossible, at least without an LLM, to avoid destroying comments and similar things.
So far, so good. But when working on PUPS, I realized that XML has a few additional security implications. XML is, when opened in the browser, executable code. It can either contain an XSLT stylesheet that will convert it into a full HTML page, or it may contain a <script> tag to execute directly. You have to be really careful when writing any XML file on a web server that is shared with web applications such as a CMS. Otherwise, when the user opens the XML in their browser, the scripts in the XML get credentials that the user has on the server.
That's why PUPS will prefer JSON and Markdown wherever possible. They just don't have the same risk. XML-based formats will be served with Sandbox-CSP and other precautions. HTML and JS will not be allowed at all.