Re: proposal: simplify prologue alternatives into %code

[Hans Aberg]

On 9 Jan 2007, at 14:36, Paolo Bonzini wrote:

> > Then a special directive %typed triggering the Bison type system  
> > may not be required, as this if %cast appears, one knows the .y  
> > file is typed. Another way would that %cast can appear, but is not  
> > used, unless a %typed also appears.
>
> %typed is unnecessary if a <foo> tag is used in the grammar.

I am aware of that: %typed would be used to ensure all stuff is typed.

> While now <foo> is always a member name, it will be a type name in  
> Java and it would be a type name also in C++ if one wants to expand
>
> %type<some_class>  xyz
>
> to {static,dynamic}_cast<type>(*var).
>
> In fact, something aking to your %typed is already exported to the  
> skeletons as b4_tag_seen_flag, where we have
>
> m4_if(b4_tag_seen_flag, 0,
> [[typedef int YYSTYPE;
> # define YYSTYPE_IS_TRIVIAL 1]]

The changes are small. But on should also keep the door for other  
langauge than C/C++/Java.

> I'm also against a new directive such as your %cast, since Bison  
> already has too many and -- in this particular case -- a simple % 
> define is all that is needed.

Well, it is a fundamental feature of Bison that needs to be expanded.  
So one adds commands to help the user. If, in principle, all commands  
could be replaced by %define, woud you want to use it?

> > The implementation of %typed would be the same as %union
>
> It would be interesting if you took a look at the implementation of  
> %union (which is just a "code muscle", equivalent to something like  
> "%code stype") before making this imperscrutable

Sorry, this is not in my dictionary.

> statement. :-)

I am not sure what you are meaning: I implemented it years ago in  
Bison, but I have just forgotten it. I jus do not remember the details.

  Hans Aberg