Re: experimental features in Bison 2.3a+

[Hans Aberg]

[I get permenant error on Help-Bison. This list eseem to work]

On 6 Dec 2006, at 10:55, Joel E. Denny wrote:

> > > For verbatim strings:
> > >
> > >   %define "NAME" "STRING"
> >
> > No, just one line. This why <code> had to be added.
>
> Do you have a specific need for multiline strings that are not  
> parsed as
> code?

I have the need for code to be added, therefore the feature was added  
to %define. But, by the reason I gave, that is, output language  
parsing, I think it is better to have a separate command %code for  
adding code. This is a conclusion I came to rather recently, some  
weeks ago. Before that, I still felt it would be better to have %code  
for code, as it would look better, but I did not really care. So I  
did not complain when it appeared in %define. But if you are going to  
add a host %code like commands anyway, it seems better to have single  
one with an argument. It will free up the Bison language name space -  
fewer keywords needed.

As for other uses of %define, I have no such immediate needs. But I  
think such needs may turn up, at least for others. Then I think it is  
best to reserve it for verbatim macros. Macros that require special  
parsing should be given special command names - that is the principle  
I see.

> > > For code that needs to be parsed:
> > >
> > >   %define "NAME" { CODE }
> > >
> > > How does calling it %code in the second case solve the problem of  
> > > multiple
> > > languages?
> >
> > Because parsing of the code part needs to be changed according to  
> > language.
> > For a macro define, that seems to be a poor practice.
>
> So the issue isn't with multiple languages?

Yes: right now, the languages that are parsed have similar parsing  
and the code additions are similar. But suppose the parsing becomes  
more different, plus that there is a need for adding verbatim code.  
Then %code should be used for regular code, and %define for the  
verbatim code.

> It's with parsing CODE at all
> when the directive is named %define?
>
> This doesn't bother me.  { CODE } is always parsed no matter where it
> appears in the Bison grammar file.

So, if %define is used to parsing code at all, I think it should not  
parse comments, strings and characters, etc..

And, I think removing the redundant quotes around NAME is pretty  
straightforward. There, I think Bison should admit '-" in macro  
names, as it will look good in grammar rules, and therefore  
elsewhere. - I do not know if M4 handles it.

  Hans Aberg