Re: %union and shared pointers to AST nodes

[Hans Aberg]

On 27 Sep 2005, at 10:40, Evan Lavelle wrote:

> Frans Englich wrote:
> > (From the top of my head it would be nice if one could combine  
> > different types for different types of "entities". Say, a union  
> > for tokens, and an YYSTYPE struct for non-terminals.

This is not possible, as all parser semantic values use the same stack.

> I personally don't see any advantage in the yystype/union/struct  
> mechanism; I would prefer to see a clean token interface between  
> flex and bison, which is completely independent of bison's $/typing  
> mechanism (which flex needs to know nothing about).

If one is using union for semantic type, some type checking is needed  
to avoid the code breaking. A similar happens with C++, if one is  
using static_cast.

The original Lex(Yacc setup, is that the parser includes the lexer  
directly, not a header. When changing this, and with additional  
complications under C++, the interfaces have to be reworked. Probably  
on a todo-list.

  Hans Aberg