RE: Minor problem with calc-quit

[Drew Adams]

    Turns out I had Drew Adams' "frame-cmds.el" in my load path, which
    contains its own version of delete-windows-on and which somehow was
    being loaded even for "emacs -Q".
    
    #> The function `delete-windows-on' returns nil when there is no *Calc
    #> Keypad* (i.e., the argument for `delete-windows-on' is nil; I tested
    #> that) and (presumably) deletes the buffer containing the key pad when
    #> it exists.
    
    Right... It turns out that version from frame-cmds.el doesn't handle
    the case of "buffer" argument being nil all that well... Drew, you
    might want to do something about that :)
    
Fixed in frame-cmds.el. Thanks for the report.

Sounds to me like calc is counting on an undocumented "feature" of 
`delete-windows-on', which might not be such a good idea.

Perhaps the doc string for `delete-windows-on' should say something about this 
behavior, to legitimize programs counting on it. It says, instead, that "BUFFER 
must be a buffer or the name of an existing buffer", which obviously excludes 
nil as a valid value. The doc string suggests that with any other value than 
those mentioned you'll get an error, or unpredictable results, or... Perhaps it 
should say explicitly that it returns nil if BUFFER is nil.