Re: Any Method of reading the token values off the stack

[Tim Zimmerlin]

Hans,

I understand that the state is embedded together with the symbol values 
on the stack; however, I do not know the memory structure. We know that 
an LALR(1) parser generated by bison supplies the $1, $2, etc. values 
from the stack somehow to the actions of the rules.

I would like some API that I can call from the actions of the rules. 
Error recovery needs only "read access" in order to output the token 
values in their proper order for later manual editing. After 
encountering an error, the parser resynchronizes (I hope). I intend to 
let the normal error recovery pop all the incorrect symbols off the 
stack to get to a recovery state.

Is there some way within the actions to determine the number of symbols 
& the symbol values? The symbol values held in the stack are lost 
during normal error recovery, leaving a hole around the error in the 
input sequence of symbols.

I hesitate to modify the stack because I hope it is not necessary. I 
hesitate to create a separate parse structure due to performance.

Thanks,
Tim Zimmerlin

On May 5, 2004, at 1:22 PM, Hans Aberg wrote:

> At 23:23 -0400 2004/05/04, Tim Zimmerlin wrote:
> > In my error recovery processing, I want to read the token values in
> > FIFO order off the stack, in yyerror(), before flushing it as part of
> > the default "error" rule processing
> >
> > Does bison provide direct stack access functions/macros? I would like
> > to know how many tokens are on the stack. I would like to read them in
> > FIFO rather than LIFO order to save them as originally read.
>
> Bison produces an LALR(1) parser, which on its stack only contains
> arbitrarily numbered LALR(1) states, represented by integral values. 
> These
> state numbers are described in the .output file that Bison writes when
> invoking the --verbose option. They represent something more 
> complicated
> than just token values: so called sets of items, where an item is
> essentially a parser rule with a dot indicating how far the parsing has
> proceeded.
>
> So the parser stack never contains any tokens. If you need to use such
> info, you must put it there yourself.
>
>   Hans Aberg