>It seems that many Big Data systems read in a dialect of JSON, with one JSON "file" per line. 

Indeed, that's a popular format, and it's an example of how standards can evolve through community evolution. Perhaps it will become popular enough that someone writes an RFC for it and allocates it a media type.

This kind of evolution has benefits and drawbacks. You often end up with a nice new capability that isn't universally supported (for example, use of non-Latin characters in email addresses), and it's a fine line whether the new capability is useful if not everyone supports it. At worst, you end up with a standard that's so woolly it's a nightmare (take CSV as an example).

XML has been very resistant to this kind of evolution. We've seen XML parsers that don't support some features in the standard (notably DTDs), but we've seen little tendency to support extensions (such as relaxing the rules on element and attribute names, or allowing the outermost element to be omitted). Perhaps this is because there are so many parsers in use and because they aren't easy to change. It certainly means that XML is a much stronger interoperability standard than some others; but it does lead to a certain amount of frustration because everyone can see that some of the rules (like disallowing nested comments) are just plain silly.

Michael Kay

Saxonica