Re: set flex start conditions from bison?

[Akim Demaille]

Le 27 août 09 à 20:52, Rui Maciel a écrit :

> > I don't know what Flex you use, but using 2.5.35 it does not compile:
> > you're using push_state, which is static in the scanner.c file, from
> > the parser.c file.
>
> Yes, that's the most recent SNAFU I stumbled on. Yet, it only goes  
> that far if I add to the flex source
> file the undocumented, unexplained #define  
> YY_HEADER_EXPORT_START_CONDITIONS, which seems like a bit of a
> hack and not a good way to go around things.

I'm not sure we refer to the same thing.  Yet, the start conditions  
are exported, but push_state and the like are not.  They are static in  
the *.c file.

> > It seems bizarre to me to play with start conditions and  
> > yy_push_state
> > from the parser.  I would rather write some wrapper function in the
> > scanner.l file, say scan_enter_element_mode(), and use that function
> > from my parser.
>
> Why is that a better alternative to simply calling yy_push_state()  
> and co from within the bison code?

Encapsulation.  It seems to me that start conditions are low-level  
details that should not escape the scanner file.

> Handling yy_push_state() directly from bison appears to be a cleaner  
> way of doing things.

Ok.

> > I agree it also seems bizarre that start conditions are defined in  
> > the
> > header file, but that yy_push_state is static.
>
> Is it possible that this is a flex bug?

I don't know, ask them.