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

[Gregory Heytings]

> This seems like its opening the door for yet another bug.

That's always possible, of course.  Jim tested it extensively.  Can you do 
the same?

> Gregory, could you please take a look at the version of pcomplete-arg 
> that I've sent?

I did.  It's not better (and in fact much more bug-prone) than reusing 
pcomplete-actual-arg.