I merely did it that way for minimality's sake -- so that there's no change of the API and programmatic use of 'cd' will continue to function as before. But I didn't succeed very well, as the cases where '-' is effective (script, etc.) show. So I guess going with always invoking that special behavior is OK.

Agreed about the message, it's short and to the point.

• libinterp/corefcn/dirfns.cc: When issuing "cd" from the top level, save

  current working dir before the change. When doing "cd -", change to that
  one. Caveat: works in a little more contexts than just from the prompt.
  (For example, anon. function defined at top level that does "cd", script
  called from top level.)

----------

This is useful when you're working in a directory with code, but temporarily want to load some data. You'd "cd" into that directory, but to return to the first one, don't need to enter it on the command line (which is inconvenient, as the completion doesn't behave properly when the partial path has more than one component.)

The idea is to make it work only when issued from the input prompt. However, the closest I found is symbol_table::at_top_level(), which presumably embodies the "workspace" concept. So, "cd -" also changes to the last WD in e.g. M-scripts called from the top level, which may not be desirable. (But then again, it's unlikely to be practically relevant -- who would call their directory "-"?)