Binary large object, and character large object (aka BLOB/CLOB) support is needed.

Rather than transforming a large binary object into hexBinary or string, instead one annotates the DFDL schema to specify that a BLOB/CLOB handle is preferred. The Infoset then contains not the data, but a token/object that can be fed to an "open" API to access a stream of the data. The token/object is a small sized thing. This allows data formats like JPEG 2000 where payloads are often gigabytes.

Note: implementing this requires revising our ByteBufferDataInputStream layer into one that does not require the input data to fit in memory.

See article on Wiki about large objects:

https://opensource.ncsa.illinois.edu/confluence/pages/viewpage.action?pageId=54427836