Re: too many warnings from Bison CVS for Pike

[Hans Aberg]

On 29 Jan 2006, at 21:56, Joel E. Denny wrote:

> On Sun, 29 Jan 2006, Hans Aberg wrote:
>
> > On 29 Jan 2006, at 20:11, Joel E. Denny wrote:
> >
> > > >   %define pre-switch { if (yylen >= 1)  $$ = $1; }
> > >
> > > This isn't yacc compatible.
> >
> > That depends on the interpretation of the Yacc specification.
>
> ???

???

> > If I look for other uses of the word "default" and the wording "by  
> > default",
> > it means "in the case no other explicit directions has been  
> > given". So "By
> > default, the value of a rule shall be the value of the first  
> > element in it."
> > means that this is the value if no other explicit value has been  
> > given, and if
> > such an explicit value has been given, the default rules does not  
> > apply.
> >
> > This is still unclear, because one can write a rule action  
> > explicitly as {}.
>
> Which doesn't give any `explicit directions' for the value for the  
> rule.

Sure it does: Do not do anything. This is useful under C++, if one  
wants to avoid unneeded copying.

> > Should then $$ = $1 be applied?
>
> Under yacc, I'm saying yes.

But it is not a question of what you think, but what the POSIX spec  
intends. This is unclear.

> > - There is a difference between explicitly
> > indicating a rule action, and explicitly assigning a value to $$.
>
> That's exactly my point.
>
> Anyway, we can debate this all week and get no where.  I think Paul's
> point is valid: regardless of what the Open Group says, executing $ 
> $ = $1
> before explicit actions will offer the widest compatibility with  
> existing
> yacc grammars.

You are jumping to conclusions: some grammars may assume this, in  
light that it is something has has worked in the past. In case of a  
change, one will have to consider the effort to rewrite such grammars.

> I say we simply state our interpretation of the spec, and
> use that interpretation as the justification for handling this with
> --yacc.

As a principle of development, this is a poor strategy: just because  
something has worked in the past against existing Bison version, but  
unofficial, it does not mean one should keep it in future versions.  
One example is the unofficial compile C as C++ that existed for  
awhile, but that was abandoned in view that it was too difficult to  
maintain.

I think that the Open Group should clarify its intent. When this has  
been done, one can take a new look at the problem. In the interim,  
one can keep the $$ = $1 executed always, as in old Bison version, as  
it does not cause any harm under C. If the code placement %define  
eventually appears in Bison, it becomes easy enough to admit a  
switch, in the case the Open Group says so, in which case it will be  
important for grammars to change as well. The C++ skeleton file  
should do what is right for C++, i.e., only execute $$ = $1 for rules  
with a default action (i.e., no action brackets {...} present).

  Hans Aberg