In most cases, removing a non-whitespace IFS character "delimits" a field in the sense that it terminates it, so the presence of such a character at the end of a field does not change its expansion:
$ IFS=X $ set -- Y; printf '<%s>' $* <Y> $ set -- YX; IFS=X; printf '<%s>' $* <Y> However, when a positional parameter ends with a non-whitespace IFS character and is followed by another positional parameter, the removal of the IFS character seems to create an additional field: $ set -- YX Y; printf '<%s>' $* <Y><><Y> Is this intentional? It sort of makes sense since $* is expanded as if it were $1c$2 (where c is the first character of IFS), i.e. the expansion above matches what is produced by set -- YX Y; v=$*; printf '<%s>' $v but that doesn't seem to respect the prescribed behavior of first producing a field for each positional parameter, and then subjecting each field to field splitting. Array behavior matches positional parameter behavior here. FWIW, this differs from the ksh behavior: $ set -- YX Y; printf '<%s>' $* <Y><Y>
