> Right, I suspect get-position and port-position relation hasn't been
> thought out wrt. buffering.  Yet,
> the read! callback of make-custom-textual-input-port allows multiple
> characters to be read, implying it allows
> character buffering.

Good point about the buffering.

Shouldn't set-position implicitly flush the buffer, and get-position
count only the characters/bytes that the user of the port has already
received from the custom read procedure?

The flushing would get tricky for ports that don't have an obvious
random-access backing store like a vector or a regular file. In
principle, we could say the port has to cache any buffered data the user
hasn't yet read, in case the user wants to seek back to the same
position later. But ports without random-access backing stores generally
do not support seeking anyway, so maybe it's the user's own problem if
they want that :)