This was an issue that we ran up against as well. The solution
that we chose was to to default to PRS if all characters
were indeed printable. On input, if an input string contains an
escaped octal character (\0###) we encode T61.
However, this assumption may not be followed by all. To properly validate
a certificate, you need to remember exactly what was used. I've decided
to dispense with these issues and to just remember the entire DER encoding!
Implementationally speaking, I can afford the memory and to decode
is not computationally expensive at all.
-Ray