Suppose field 1 is 3 bits long.
Suppose field 2 is a string in utf-8. dfdl:emptyValueDelimiterPolicy='none'
If field2 has any contents, then those characters must begin on a byte boundary, so before the first character we will skip 5 bits.
What if field2 is of length 0 ? Do we skip 5 bits anyway or not?
Note that if dfdl:emptyValueDelimiterPolicy is "both", and an initator or terminator is defined, then even if length is 0 we definitely will unparse a character at least so the 5 bits of alignment are needed in that case.
But if there is no framing, and no content, do we align to the text mandatory alignment or not?
Now, I believe the answer to this should be "yes" we skip the 5 bits anyway. Because if field 2 is lengthKind='pattern', then when parsing, we have to scan for a match to the pattern, and so we have to know where to start the scan, which has to be at a byte boundary. So even if there is no match, so the length at parse time is zero, we still had to skip 5 bits before we could start scanning.
So to be consistent with this, I believe unparsing must also output the 5 bits regardless of whether the string is length 0 or not.
So there's what should it do, but also if all the constructs for this work in IBM DFDL, I'm curious what it actually does.