Re: no $$ in %destructor = no unset value warnings?

[Joel E. Denny]

On Thu, 14 Sep 2006, Akim Demaille wrote:

> >>> "Joel" == Joel E Denny <address@hidden> writes:
> 
>  > a. value is likely set/constructed = $$ appears in semantic action
>  > b. value is likely destroyed = $$ appears in %destructor
> 
>  > That is, I think Bison shouldn't warn about !a when !b.  However,
>  > Bison currently considers a value to be likely destroyed by a
>  > %destructor regardless of whether $$ appears in the %destructor.  I
>  > think that's inconsistent.
> 
> It is inconsistent, you are right, but I would have chosen the other
> possibility: just like we invite users to YYUSE (or whatever the
> macro) in rule actions, we should do that in destructors and printers.
> Not because it is always wrong not to use the value, but because it's
> always possibly an error.

Ok, I think I agree.

>  > Moreover, it might be convenient to override a per-type or default
>  > %destructor by defining an empty per-symbol %destructor, which
>  > would suppress the warnings:
> 
>  >   %destructor { free ($$); }
>  >   %destructor { } symbol_with_no_value
>
>  > What do you think?
> 
> Err?  Default destructors apply whether there is a semantic value or
> not?  I have plenty of catch-up to do :(

I was thinking of a user-defined YYSTYPE (no %union) with no <...> usage.  
In this case, Bison would have no idea what symbols truly have semantic 
values.  So, I think we need the ability to remove the destructor for 
specific symbols.  There might even be cases where nearly all symbols of a 
particular type need a destructor, but a few don't.  Finally, I still like 
the idea of a %destructor that doesn't access $$ but perhaps just prints a 
message, and I don't think Bison should warn about rules that don't use 
such a $$/$n.

You once discussed the possibility of named semantic values in grammar 
rules:

  exp(sum): exp(term1) '+' exp(term2) {
    $sum = $term1 + $term2
  }

And instead of USE($1), you suggested empty parens, for which I still 
prefer (!).  I think empty parens make a nice default value name:

  grammar(): defs() rules() epilogue(!) {
    $grammar = new_grammar( $defs, $rules );
  }

Maybe something similar is useful with %destructor/%printer:

  %destructor(!) { printf ("A SYM was discarded.\n"); } SYM

Because the user has explicitly declared that SYM does not need to destroy 
its value, Bison would not warn about unused SYM values in %destructor or 
semantic actions.

By the way, shouldn't Bison warn when there's %union or <...> usage in the 
grammar but a $$ without a type appears in a %destructor/%printer?  This 
is already the case for semantic actions.