DFDL-893 but worth carving off as a lower-priority aspect.
This was a comment on the original January 2011 version 1.0.3 spec of DFDL.
The Sept 2013 1.0.4 spec says:
When utf16Width is 'variable', then on parsing an un-paired surrogate codepoint causes a decode error, which can be controlled via dfdl:encodingErrorPolicy.
Is this the actual behavior, i.e., does ICU let us even do this? (If not then this will be an issue with the DFDL spec.)