Re: %destructor feedback

[Akim Demaille]

Le 4 janv. 06 à 20:29, Joel E. Denny a écrit :

> On Wed, 4 Jan 2006, Frank Heckenbach wrote:
>
> > Joel E. Denny wrote:
> >
> > > Yikes.  Specifying a meaningless type is ugly.
> >
> > Indeed, it looks ugly. OTOH, this seems a rather uncommon situation
> > (using $-n is already ugly, IMHO -- is it POSIX conformant
> > actually?), so I'm not convinced we need special effort in Bison to
> > spare a few characters in such a particular situation. (But of
> > course, it's not up to me to decide how much effort others put in
> > it. ;-)
>
> I don't see $<foo>3 as simply a waste of characters.  I see it as
> misleading.  It begs the question, "What does `foo' mean?".

Well, I wrote foo, but I should have written the actual member name:
$<val>3.  That's what is needed anyway in the action itself when
assigning to $$ (well, $<val>$).

> > > It looks strange at first glance, but the first thing that pops  
> > > into my
> > > mind is:
> > >
> > >   $<>1; $<>2; $<>3; $<>4; $<>5;

Currently this is a syntax error: you are reminded there should be
something there.

> > In this case, Bison could also warn if such a variable is actually
> > used, e.g.:
> >
> >   $<>1;
> >   foo ($1);  /* or: foo ($<type>1); */
>
> Yes, this is nice.  That is, if you have a complex action, you might
> accidentally declare $<>1 and use $1 too.  That hints that there  
> might be
> something wrong with the logic in your action, so it's nice to have  
> bison
> warn you.  I don't think there would have been any easy way to  
> detect this
> mistake with YYUSE_VAL.  $<>1 is indeed more flexible.

I'm not convinced we need that much.  I can leave with $<val>3 :)