Re: quick bison Q

[David Durham]

Paul Eggert wrote:

> > Date: Fri, 06 Dec 2002 09:29:19 -0600
> > From: David Durham <address@hidden>
> > User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.1) Gecko/20020826
> >
> > I was wondering why
> >
> > %union foo
> > {
> > }
> >
> >
> > doesn't work any more?  It doesn't like 'foo'
> >    
>
> The Bison reader was redone recently, and my guess is that we didn't
> know about this undocumented feature.  Also, POSIX does not require
> support for this feature.  Instead of using "%union foo {...}", you
> are supposed to do something like this:
>
>  %{
>    union foo { int ival;};
>    #define YYSTYPE union foo
>  %}
>  
ahh.. I'll do it that way

>  
>
> > the reason I need to name it is to be able to declare it as a friend
> > of a C++ class.
> >    
>
> Perhaps we should add this feature back to Bison, but I'd like to know
> exactly what is needed.
>  
Yeah.. I was surprised that mandrake is still packaging 1.35.. I suppose 
because they've had problems using 1.5/7 with compiling some things.. 
might want to ask the bison packagers for mandrake: address@hidden

> In C, the only thing that can appear between the "union" and the "{"
> is an optional identifier (also white space and comments, of course).
> Is the same true for C++?
>  
I just checked the C++ spec, and the grammar itself shows union to have 
the exact syntax as a class.  However as far as I know, there is no 
meaning for inheritance of unions (which is really the only specifier 
that can come after the ident).

However C++ does allow union to be followed by some template keyword 
stuff and some allows idents with '::' between them. I really don't 
think that bison should have to worry about what they mean.  It could 
just find what's between the union keyword and the {, remove all the 
whitespace from that sequence of chars and define that as the character 
sequence of the typename for the union.   If bison places that 
everywhere in the C code where the typename of the union is needed then 
when a C++ compiler compiles it it should, I think, work fine.

On the otherhand, you could check what bison 1.35 allowed and go with 
that.   I think it just allowed a simple ident after the union.  


> I have a language-lawyer question for you, since I don't use C++
> myself.  In C++, what exactly can appear between the "union" and the
> "{"?  Is it OK for Bison to restrict it to be a valid C identifier?
> Here is the grammar for a valid C identifier:
>
>  identifier = letter (letter | digit)*
>
>  digit = [0-9]
>  letter = [_a-zA-Z] | "\u" x x x x | "\U" x x x x x x x x | idl
>  x = [0-9a-fA-F]
>  idl = <implementation-defined letter>
>
> A backslash-newline can be inserted between any two characters, but I
> don't think Bison should have to worry about that.
>
> Those "implementation-defined letters" worry me a bit, though, since
> there's no easy way for Bison to discover whether a letter is allowed
> by the implementation.
>
>  
From looking at the grammar C++ idents are exactly the same as C 
idents. And the \-newline stuff I believe would be handed by the 
preprocessor and not the C/C++ parser itself.  And if something works in 
C it is supposed to work in C++ (perhaps with compiler warnings, but no 
errors)

I'm not sure what is meant by implementation-defined letter?  Is that 
like a non-US-english character?  Or is it something that can be defined 
by the user at bison-time?