bug#60464: 29.0.60; Regression - pcomplete-arg fails with argument 'last

[Gregory Heytings]

> >                    (_      (- pcomplete-index (or index 0))))
> >                  (or offset 0))
> >                pcomplete-args)))
> > -    (if (stringp arg)
> > +    (if (or (stringp arg)
> > +            (eq index 'last))
> >          arg
> >        (propertize
> > -       (buffer-substring (pcomplete-begin index offset)
> > -                         (pcomplete-begin (1- (or index 0)) offset))
> > +       (car (split-string (pcomplete-actual-arg index offset)))
> >         'pcomplete-arg-value arg))))
>
> I'm not sure what specific problem this is trying to solve (is it the 
> choice of the "index" or is it the precise buffer positions of the 
> bounds)?

I admit I don't understand your question.  It tries to use 
pcomplete-actual-arg, which uses buffer-substring, to get the text 
representation of the argument.  It's not guaranteed to work in all cases, 
though, because it simply assumes that arguments are separated by spaces. 
But it should be "good enough".

The (eq index 'last) means that when index is 'last we return the 
argument, even when it is not a string (or more precisely: when it is a 
list of strings, and the code assumes that a non-string arg is a list of 
string), because that's in fact what the caller expects (or at least 
that's what Eshell expects) in that case: that list is displayed in 
*Completions*.