Re: too many warnings from Bison CVS for Pike

[Hans Aberg]

On 22 Feb 2006, at 01:57, Joel E. Denny wrote:

> > Then the semantic type is a non-POD class
>
> OK, so you #define YYSTYPE, I take it.  It seems that this usage is
> addressed in "lalr1.cc", which I should have checked before.

I use a modification of an older lalrl.cc, which in its turn Akim  
used part of to modify the new one. So I do not expect there will be  
any problem with non-POD types as such. The problem I have is that I  
use (together with Flex) a traditional C++ header/source file setup,  
for which I need a variation of %define, which can place the code.

The problem is when using %union, which implements a 'union', which  
cannot be used with a POD type, according to the C++ standard. So  
here one needs a variation of %union that instead admits one to  
define suitable macros. Then there are two problems: First, the  
fields of a 'union' are used by selecting a union field 'y' in 'x' by  
the syntax 'x.y'; in C++, this should be replaced by something  
(depending on the setup) involving dynamic_cast or static_cast. This  
should not be difficult to handle using macros. In addition, Bison  
assumes that the default actions can be computed using $$ = $1. Now,  
in C++, using a non-POD type, this may not work due to the strong  
typing; instead one gets $type>$ = $<type>1. This one, I do not yet  
know if it can be reduced to the existing Bison implementation, or if  
one, in the parser 'switch' statement, needs to have a row of default  
actions depending on type.

  Hans Aberg